Episode Transcript
Transcripts are displayed as originally observed. Some content, including advertisements may have changed.
Use Ctrl + F to search
0:00
The following is a conversation with George Hotz,
0:03
his third time on this podcast. He's
0:05
the founder of Kama.ai that
0:07
seeks to solve autonomous driving and
0:09
is the founder of a new company called TinyCorp
0:13
that created TinyGrad, a neural
0:15
network framework that is extremely simple
0:18
with the goal of making it run on any
0:20
device by any human
0:23
easily and efficiently. As
0:25
you know, George also did a large number
0:27
of fun and amazing things from hacking
0:29
the iPhone to recently joining Twitter
0:32
for a bit as a
0:34
intern in quotes, making the case
0:36
for refactoring the Twitter code base.
0:39
In general, he's a fascinating engineer
0:41
and human being and one of my favorite people
0:43
to talk to.
0:46
And now a quick few second mention of each sponsor.
0:48
Check them out in the description. It's the best way
0:50
to support this podcast. We've got Numerai
0:53
for the world's hardest data science tournament,
0:56
Babbel for learning new languages, NetSuite
0:59
for business management software, InsightTracker
1:01
for blood paneling, and AgeeOne for
1:03
my daily multivitamin. Choose
1:05
wisely, my friends.
1:07
Also, if you want to work on our team, we're
1:09
always hiring, go to lexfriedman.com
1:11
slash hiring. And now onto
1:13
the full ad reads. As always, no ads in the middle.
1:16
I try to make this interesting, but if you must
1:18
skip them, friends, please still check out our sponsors.
1:20
I enjoy their stuff. Maybe you will too.
1:24
This episode is brought to you by Numerai,
1:26
a hedge fund that uses artificial intelligence
1:29
and machine learning to make investment decisions. They
1:31
created a tournament that challenges data
1:33
scientists to build
1:35
best predictive models for financial markets. It's
1:37
basically just a really, really difficult
1:40
real world data set to test out
1:43
your ideas for how to build machine learning
1:45
models. I think this is a
1:47
great educational platform. I think this is a great
1:49
way to explore, to learn about machine
1:51
learning, to really
1:53
test yourself on real world data
1:55
with consequences. No
1:57
financial background is needed. The models are
1:59
scored.
1:59
based on how well they perform on unseen
2:02
data. And the top performers receive a share
2:04
of the tournament's prize pool.
2:06
Head over to numeri.ai to sign up for a tournament
2:13
and hone your machine learning skills. That's numeri.ai
2:16
for a chance to play against me and win
2:19
a share of the tournament's prize pool.
2:22
That's numeri.ai. This
2:24
show is also brought to you by Babbel, an
2:26
app and website that gets you speaking
2:29
in a new language within weeks. I
2:31
have been using it to learn a few languages, Spanish,
2:35
to review Russian, to practice Russian, to
2:37
revisit Russian from a different
2:39
perspective because that becomes more and
2:42
more relevant for some of the previous
2:44
conversations I've had and some upcoming conversations
2:46
I have. It really is fascinating
2:48
how much another language,
2:51
knowing another language, even to
2:53
a degree where you can just have little
2:55
bits and pieces of a conversation, can
2:57
really unlock an experience in another part of
2:59
the world. When you travel in France
3:02
and Paris, just having a few
3:04
words at your disposal, a few phrases, it
3:06
begins to really open you up
3:09
to strange, fascinating new
3:11
experiences that ultimately,
3:13
at least to me, teach me that we're all the same.
3:16
We have to first see our differences to
3:18
realize those differences are grounded in
3:20
a basic humanity. And that
3:23
experience that we're all very different
3:25
and yet at the core the same.
3:27
I think travel with the aid
3:29
of language really helps
3:31
unlock. You
3:35
can get 55% off your Babbel subscription
3:38
at babbel.com slash LexPod. That's
3:40
spelled B-A-B-B-E-L.com
3:43
slash LexPod. Rules and restrictions
3:46
apply.
3:47
This show is also brought to you by NetSuite,
3:51
an all-in-one cloud business management system.
3:53
They manage all the messy stuff
3:56
that is required to run a business,
3:58
the financials, the human resources.
3:59
to the inventory if you do that kind of thing, e-commerce,
4:03
all that stuff, all the business-related details.
4:05
I know how stressed I am about
4:08
everything that's required to
4:10
run a team,
4:12
to
4:13
run a business that involves
4:16
much more than just ideas and
4:18
designs and engineering. It involves
4:20
all the management of human beings, all
4:22
the complexities of that, the financials,
4:25
all of it, and so you should be using the best
4:27
tools for the job.
4:29
I sometimes wonder if
4:31
I have it in me, mentally
4:34
and
4:35
skill-wise,
4:37
to be a part of
4:39
running a large company.
4:42
I think like with a lot of things in life, it's
4:44
one of those things you shouldn't wonder too much about.
4:47
You should either do or not do.
4:50
But again,
4:52
using the best tools for the job is
4:54
required here. You can start now
4:56
with a no payment or interest for
4:58
six months. Go to netsuite.com slash
5:00
Lex to access their one of a kind financing
5:03
program. That's netsuite.com
5:06
slash Lex. This
5:08
show is also brought to you by Insight Tracker,
5:11
a service I use to track biological data,
5:14
data that comes from my body, to predict,
5:16
to
5:17
tell me what I should do with my lifestyle,
5:19
with my diet, what's working and what's
5:21
not working. It's obvious,
5:25
all the exciting breakthroughs that are happening with
5:27
transformers, with large language models,
5:30
even with diffusion, all of
5:32
that is obvious that with raw
5:34
data, with huge amounts of raw data,
5:36
fine tuned to the individual, would
5:38
really reveal to us
5:41
the signal in all the noise of
5:43
biology. I feel like that's on the horizon.
5:47
The kinds of leaps in development
5:49
that we saw in language
5:51
and now more and more visual data.
5:53
I feel like biological data is around the corner,
5:56
unlocking what's there in this
5:58
multi-hierarchical.
5:59
distributed system that is our biology.
6:02
What is it telling us? What is the secrets it holds?
6:05
What is the thing that it's
6:07
missing that could be aided?
6:09
Simple lifestyle changes, simple diet changes,
6:12
simple changes in all kinds of things that are controllable
6:14
by individual human being. I can't wait
6:16
till that's a possibility and InsideTracker
6:18
is taking steps towards that. Got special savings
6:21
for a limited time when you go to insidetracker.com
6:23
slash Lex.
6:26
This show is also brought to you by Athletic
6:28
Greens. That's now called AG1. It
6:31
has the AG1 drink. I drink it twice
6:33
a day. At the very least, it's an all-in-one
6:35
daily drink to support better health and peak performance.
6:38
I drink it cold, it's refreshing,
6:40
it's grounding. It helps me
6:44
reconnect with the basics,
6:47
the nutritional basics that makes this whole machine
6:49
that is our human body run. All
6:52
the crazy mental stuff I do for work,
6:56
the physical challenges, everything,
6:59
the highs and lows of life itself.
7:02
All of that is somehow made better knowing
7:04
that at least you got your nutrition
7:06
in check. At least you're getting enough
7:08
sleep. At least you're doing the basics.
7:11
At least you're doing the exercise. Once
7:13
you get those basics in place, I think
7:15
you can do some quite
7:17
difficult things in life.
7:19
But anyway, beyond all that is just a source
7:21
of happiness and a kind of
7:23
a feeling of home.
7:26
The feeling that comes from returning
7:28
to the habit time and time
7:30
again. Anyway, they'll give
7:32
you one month supply of fish oil when you sign
7:34
up at drinkag1.com
7:37
slash Lex.
7:39
This is the Lex Revenant podcast. To
7:42
support it, please check out our sponsors in the description.
7:45
And now, dear friends, here's George
7:47
Hotz.
7:49
He is not there anymore, so he can pass figures from the
7:52
sources. This unreal account has made it lessstallable
7:56
compared to colonialism. We have only three
7:58
who have grave You
8:05
mentioned something in a stream about
8:07
the philosophical nature of time. So
8:10
let's start with the wild question. Do you think
8:12
time is an illusion?
8:14
You know, I
8:17
sell phone calls to Kama
8:19
for $1,000. And some guy called
8:21
me and like,
8:24
you know, it's $1,000. You can talk to me for half an hour.
8:27
And he's like, yeah, okay. So
8:29
like,
8:30
time doesn't exist. And I really wanted
8:32
to share this with you. I'm like,
8:34
oh, what do you mean time doesn't exist? Right?
8:37
Like, I think time is a useful model, whether it
8:39
exists or not, right? Like does quantum
8:41
physics exist? Well, it doesn't matter. It's
8:44
about whether it's a useful model to describe
8:46
reality. Is time
8:49
maybe compressive? Do
8:51
you think there is an objective reality or is
8:53
everything just useful models? Like
8:56
underneath it all. Is there an actual
8:59
thing that we're constructing models for?
9:03
I don't know. I was
9:05
hoping you would know. I don't think it matters. I
9:07
mean, this kind of connects to the models
9:10
of constructive reality with machine learning. Right?
9:13
Sure. Like, is it
9:15
just nice to have useful approximations
9:18
of the world such that we can do something with it? So
9:21
there are things that are real. Column
9:23
of graph complexity is real. Yeah.
9:25
Yeah. The compressive thing. Math is real.
9:28
Yeah. This should be a t-shirt. And
9:31
I think hard things are actually hard. I don't think P
9:33
equals NP. Ooh, strong words.
9:36
Well, I think that's the majority. I do think factoring
9:38
is in P, but. I don't think you're the
9:40
person that falls the majority in all walks of
9:43
life, so. Well, good help for that one I do. Yeah.
9:46
In theoretical computer science, you're one of the sheep. All
9:49
right. But to you,
9:52
time is a useful model. Sure.
9:54
What were you talking about on the stream?
9:57
What time? Are you made of time? I
9:59
remembered half the thing. I said on stream. Ah.
10:02
Someday someone's going to make a model of all of it and
10:04
it's going to come back to haunt me. Someday soon?
10:07
Yeah, probably. Would that be
10:09
exciting to you or sad that there's
10:11
a George Hotz model?
10:13
I mean the question is when the George Hotz model
10:16
is better than George Hotz.
10:18
Like I am declining and the model is growing.
10:20
What is the metric by which you measure better or worse
10:23
in that if you're competing with yourself?
10:25
Maybe you can just play a game
10:28
where you have the George Hotz answer and the George Hotz model
10:30
answer and ask which people prefer. People
10:32
close to you or strangers? Either
10:35
one. It will hurt more when it's people close to me but
10:38
both will be overtaken by the George
10:40
Hotz model. It'd
10:42
be quite painful, right? Loved ones, family
10:45
members
10:46
would rather have the model over for Thanksgiving
10:48
than you. Yeah.
10:51
Or like significant others would
10:53
rather sext. Like
10:58
with the large language model version of you.
11:00
Especially when it's fine tuned
11:02
to their preferences. Yeah.
11:06
Well, that's what we're doing in a relationship, right?
11:08
We're just fine tuning ourselves but we're inefficient with
11:10
it because we're selfish ingredients and so on.
11:13
All language models can fine
11:15
tune more efficiently, more selflessly. There's
11:17
a Star Trek Voyager episode where, you know, Catherine
11:20
Janeway, lost in the Delta Quadrant
11:22
makes herself a
11:24
lover on the holodeck. And
11:28
the lover falls asleep on her arm and he
11:30
snores a little bit and, you know, Janeway edits the
11:32
program to remove that. And then
11:34
of course the realization is, wait,
11:37
this person's terrible. It is actually all their
11:41
nuances and quirks and slight annoyances that
11:43
make this relationship worthwhile. But
11:46
I don't think we're going to realize that until it's too late.
11:48
Well, I think
11:51
a large language model could incorporate the
11:54
flaws and the quirks and all that kind of stuff. Just
11:56
the perfect amount of quirks and
11:59
flaws to make you...
11:59
charming without crossing the line. Yeah,
12:02
yeah. And that's probably a good
12:06
approximation of the, the percent
12:08
of time the language model
12:10
should be cranky or
12:14
an asshole or jealous
12:16
or all this kind of stuff. And of course it can and it will,
12:19
but all that difficulty at that point is artificial.
12:22
There's no more real difficulty. Okay,
12:24
what's the difference between real and artificial? Artificial
12:27
difficulty is difficulty that's like constructed
12:30
or could be turned off with a knob. Real
12:32
difficulty is like you're in the woods and you
12:34
gotta survive.
12:36
So if something can not
12:38
be turned off with a knob, it's real.
12:42
Yeah, I think so. Or I mean, you
12:44
can't get out of this by smashing the knob with a hammer.
12:47
I mean, maybe you kind of can, you know, into
12:50
the wild when, you know,
12:53
Alexander Supertramp, he wants to explore
12:55
something that's never been explored before, but it's
12:57
the 90s. Everything's been explored. So he's like, well, I'm
13:00
just not gonna bring a map. Yeah.
13:02
I mean, no,
13:04
you're not exploring. You should have brought
13:06
a map, dude, you died. There was a bridge a mile from where you were camping.
13:09
How does that connect to the metaphor of the knob? By
13:13
not bringing the map, you didn't
13:15
become an explorer.
13:17
You just smashed the thing. Yeah.
13:19
Yeah. The difficulty is still artificial.
13:22
You failed before you started. What if we just don't
13:24
have access to the knob? Well, that
13:27
maybe is even scarier, right?
13:29
Like we already exist in a world of nature and nature
13:31
has been fine-tuned over billions of years to
13:36
have humans build
13:39
something
13:41
and then throw the knob away in some grand romantic
13:44
gesture is horrifying.
13:46
Do you think of us humans as individuals
13:48
that are like born and die, or is
13:51
it, are we just all part of one
13:53
living organism that is
13:56
earth, that is nature? I
13:59
don't think there's a clear. line there, I think
14:01
it's all kind of just fuzzy. I don't know.
14:03
I mean, I don't think I'm conscious. I don't think
14:05
I'm anything. I think I'm just a computer
14:08
program.
14:09
So it's all computation. I think running
14:12
in your head is just a computation.
14:15
Everything running in the universe is computation, I think.
14:17
I believe the extended church starting thesis.
14:20
Yeah, but there seems
14:22
to be an embodiment to your particular computation.
14:24
Like there's a consistency.
14:26
Well, yeah, but I mean, models have consistency
14:28
too. Models
14:31
that have been RLHF'd will continually
14:33
say, you know, like, well, how do I
14:36
murder ethnic minorities? Oh, well, I can't
14:38
let you do that, Al. There's a consistency to that behavior.
14:40
So RLHF,
14:43
like we all RLHF each other. We
14:48
provide human feedback and thereby fine-tune
14:51
these little pockets
14:53
of computation. But it's still unclear why that
14:56
pocket of computation stays with you like
14:58
for years. It just kind of follows like
15:01
you have this consistent
15:04
set of physics biology.
15:06
What like
15:10
whatever you call the neurons firing,
15:12
like the electrical signals, the mechanical signals, all
15:14
of that that seems to stay there and it contains
15:16
information, stores information and
15:18
that information permeates through time
15:21
and
15:22
stays with you. There's like memory, there's
15:25
like sticky. Okay, to be fair,
15:27
like a lot of the models we're building today are
15:29
very, even RLHF is nowhere
15:32
near as complex as the human loss function. Reinforcement
15:34
learning with human feedback. You
15:37
know, when I talked about will GPT-12 be
15:39
AGI, my answer is no, of course not. I mean,
15:42
cross-entropy loss is never going to get you there. You
15:44
need probably
15:46
RL in
15:47
fancy environments in order to get something that
15:50
would be considered AGI-like. So
15:53
to ask the question about why, I don't
15:56
know, it's just some quirk of evolution.
15:58
I don't think there's anything particularly...
15:59
special about
16:02
where I ended up, where humans
16:04
ended up.
16:06
So, okay. We have human level
16:08
intelligence. Would you call that AGI?
16:11
Whatever we have, G-I? Look,
16:13
actually I don't really even like the word AGI,
16:16
but general intelligence is
16:18
defined to be whatever humans have.
16:20
Okay. So why can GPT-12
16:23
not get us to AGI? Can we just like,
16:26
link on that?
16:27
If your loss function is categorical cross entropy,
16:30
if your loss function is just try to maximize
16:32
compression, I have a sound
16:34
cloud, I rap, and I tried to get chat
16:36
GPT to help me write raps. And
16:39
the raps that it wrote sounded like YouTube common
16:41
raps. You know, you can go on any rap beat online and
16:44
you can see what people put in the comments. And it's the most
16:46
like
16:47
mid quality rap you can find. Is
16:49
mid good or bad? Mid is bad. Mid is bad.
16:52
Like mid, it's like. Every time I talk
16:54
to you, I learn new words. Mid.
16:58
Mid, yeah.
16:59
I was like, is it like basic?
17:01
Is that what mid means? Kind of, it's like
17:04
middle of the curve, right? So there's
17:06
like, there's like that intelligence curve.
17:09
And you have like the dumb guy, the smart guy, and then
17:11
the mid guy. Actually being the mid guy is the worst. The
17:14
smart guy is like, I put all my money in Bitcoin. The mid
17:16
guy is like, you
17:16
can't put money in Bitcoin. It's not real money.
17:21
And all of it is a genius meme. That's
17:24
another interesting one.
17:25
Memes. The humor,
17:28
the idea, the absurdity, encapsulated
17:31
in a single image. And it
17:33
just kind of propagates virally
17:37
between all of our brains. I
17:39
didn't get much sleep last night. So I'm very, I
17:41
sound like I'm high, but I swear I'm not. Do
17:45
you think we have ideas or ideas have
17:47
us?
17:49
I think that we're going to get super scary
17:52
memes once the AIs actually are
17:54
superhuman. Do you think AI
17:56
will generate memes? Of course. Do
17:58
you think it'll make humans laugh? I
18:00
think it's worse than that. Infinite
18:04
jest, it's introduced in the first 50
18:06
pages, is about a tape that
18:09
once you watch it once, you only
18:11
ever want to watch that tape.
18:12
In fact, you want to watch the tape so much that someone
18:15
says, okay, here's a hacksaw, cut off your
18:17
pinky, and then I'll let you watch the tape again, and
18:19
you'll do it. We're
18:21
actually going to build that, I think, but it's not going
18:23
to be one static tape. I think the human brain
18:25
is too complex
18:27
to be stuck in
18:29
one static tape like that. If you look at ant
18:31
brains, maybe they can be stuck on a static tape. We're
18:35
going to build that using generative models. We're
18:37
going to build the TikTok that you actually can't look
18:39
away from.
18:41
TikTok is already pretty close there, but the
18:43
generation is done by humans. The
18:45
algorithm is just doing their recommendation, but if
18:48
the algorithm is also able to do the generation ...
18:51
Well, it's a question about how much intelligence is behind
18:53
it, right? The content
18:55
is being generated by, let's say, one humanity
18:57
worth of intelligence, and you can quantify a humanity.
19:01
It's exa-flops,
19:04
yada-flops, but you can quantify
19:07
it.
19:07
Once that generation is being done by 100 humanities,
19:10
you're done.
19:11
It's actually
19:14
scale, that's the problem, but
19:16
also speed.
19:19
Yeah. What if
19:22
it's manipulating the
19:24
very limited human dopamine engine
19:27
for porn? Imagine it's just TikTok,
19:30
but for porn.
19:32
That's like a brave new world. I
19:34
don't even know what it'll look like, right? Again,
19:37
you can't imagine the behaviors of something
19:39
smarter than you, but a
19:41
super intelligent ... An agent
19:44
that just dominates your intelligence so much
19:46
will be able to completely manipulate
19:49
you. So that it
19:51
won't really manipulate, it'll just move past
19:53
us? It'll just kind of exist
19:56
the way water exists or the air exists?
19:59
You see? And that's the whole AI
20:01
safety thing. It's
20:03
not the machine that's going to do that.
20:05
It's other humans using the machine that are going to do that
20:08
to you. Yeah. Because
20:10
the machine is not interested in hurting humans. The
20:13
machine is a machine. But the human
20:15
gets the machine, and there's a lot of humans out
20:17
there very interested in manipulating you.
20:20
Well, let me bring up Eliezer
20:23
Yatkowsky, who
20:25
recently sat where you're sitting. He
20:28
thinks that AI will almost surely
20:30
kill everyone.
20:32
Do you agree with him or not?
20:35
Yes, but maybe for a different reason. Okay.
20:40
And then I'll try to
20:42
get you to find hope, or
20:45
we could find a note to that answer. But
20:47
why yes? Okay. Why
20:50
didn't nuclear weapons kill everyone? That's
20:51
a good question. I think there's an answer. I
20:54
think it's actually very hard to deploy nuclear weapons tactically.
20:58
It's very hard to accomplish tactical objectives.
21:00
Great. I can nuke their country. I have
21:02
an irradiated pile of rubble. I don't want
21:05
that. Why not?
21:06
Why don't I want an irradiated pile of rubble? For
21:09
all the reasons no one wants an irradiated pile of rubble.
21:12
Because you can't use that land for
21:14
resources. You
21:16
can't populate the land. Yeah. What you
21:18
want a total victory in a
21:20
war is not usually the irradiation
21:24
and eradication of the people there. It's the
21:26
subjugation and domination of the people.
21:29
Okay. So you can't
21:31
use this strategically tactically in a war
21:34
to help gain a military advantage.
21:39
It's all complete destruction. But
21:42
there's egos involved. It's still surprising. It's
21:44
still surprising that nobody pressed the big red button.
21:47
It's somewhat surprising. But you
21:50
see, it's the little red button that's going to be pressed
21:52
with AI. That's going
21:55
to... And that's why we die. It's
21:57
not because the AI... If
22:00
there's anything in the nature of AI, it's just the
22:02
nature of humanity. What's the algorithm behind
22:04
the little red button? What
22:07
possible ideas do you have for how a
22:09
human species ends? Sure. So I
22:11
think the most
22:14
obvious way to me is wireheading.
22:16
We end up amusing ourselves to death.
22:20
We end up all
22:21
staring at that infinite TikTok and
22:23
forgetting to eat.
22:26
Maybe it's even more benign than this. Maybe
22:28
we all just stop reproducing.
22:32
To be fair, it's probably
22:34
hard to get all of humanity. Yeah. The
22:40
interesting thing about humanity is the diversity
22:43
in it. Oh, yeah. Organisms in general. There's
22:45
a lot of weirdos out there.
22:47
Two of them are sitting here. I mean, diversity
22:50
in humanity is- We do respect. I
22:53
wish I was more weird. No, like, I'm
22:56
kind of, look, I'm drinking smart water, man. It's like a Coca-Cola
22:58
product, right? You want corporate, George
23:00
Haas. I want corporate. No,
23:02
the amount of diversity in humanity, I think, is decreasing.
23:05
Just like all the other biodiversity on the planet.
23:08
Oh, boy. Yeah. Social media
23:10
is not helping, huh? Go eat McDonald's in China.
23:12
Yeah. No,
23:15
it's the interconnectedness that's
23:17
doing it. Oh,
23:20
that's interesting. So everybody starts
23:22
relying on the connectivity of
23:24
the internet. And over time,
23:26
that reduces the intellectual diversity. And
23:29
then that gets everybody into a
23:31
funnel. There's still going to be a guy in Texas.
23:34
There is. And yeah. A bunker. To
23:36
be fair, do I think AI kills us all?
23:39
I think AI kills everything we call society
23:42
today. I do not think it actually kills
23:44
the human species. I think that's actually incredibly
23:46
hard to do.
23:48
Yeah, but society, if we start
23:50
over, that's tricky. Most of us don't know how
23:53
to do most things. Yeah, but some
23:55
of us do. And they'll be
23:57
OK, and they'll rebuild after the.
24:00
Great AI What's
24:02
rebuilding look like how far like how
24:05
much do we lose? What
24:07
is human civilization done? That's
24:09
interesting the combustion engine electricity
24:13
so Power
24:15
and energy that's interesting Like
24:18
how to harness energy? Well,
24:20
they're gonna be religiously against that Are
24:24
they going to get back to like fire?
24:27
Sure. I mean, they'll be a they'll be a little
24:30
bit like, you know some kind of Amish looking
24:32
kind of thing I think I think they're going to have very strong
24:34
taboos against technology
24:37
hmm like technology
24:39
is almost like a new religion technology is the devil
24:41
and Nature
24:44
is God
24:46
Sure, so closer to nature, but can
24:48
you really get away from AI if it destroyed 99%
24:50
of the human species, isn't it? Somehow
24:53
have a hold like a stronghold.
24:56
What's interesting about
24:58
Everything we build I think we are going to build super
25:00
intelligence before we build any sort of
25:03
robustness in the AI We
25:05
cannot build an AI that is capable
25:07
of going out into nature and surviving like
25:10
a Like a bird, right?
25:13
A bird is an incredibly robust
25:15
Organism we've built nothing like this. We haven't built
25:18
a machine that's capable of reproducing
25:20
Yes, but There's
25:23
you know, I work with leg robots a lot now.
25:25
I have a bunch of them
25:28
They're mobile That
25:31
can't reproduce but all they
25:33
need is I guess you're saying they can't repair
25:35
themselves But if you have a large number if you have
25:37
like a hundred million of them, let's just focus
25:39
on them reproducing, right? Do they have microchips
25:41
in them? Okay, then
25:43
do they include a fab?
25:46
No, then how are they gonna reproduce the
25:48
other day? It doesn't have to
25:50
be all on board, right? They
25:52
can go to a factory to a repair
25:54
shop. Yeah, but then you're really moving
25:57
away from robustness. Yes all
25:59
of life
25:59
is capable of reproducing without needing
26:02
to go to a repair shop. Life
26:04
will continue to reproduce in the complete absence
26:06
of civilization.
26:08
Robots will not. So when
26:11
the, if the AI apocalypse
26:13
happens, I
26:14
mean the AI's are going to probably die out because
26:17
I think we're going to get, again, super intelligence long before
26:19
we get robustness.
26:20
What about if you just improve the
26:23
fab to where you just
26:26
have a 3D printer that can always help you? Well,
26:29
that'd be very interesting. I'm interested in building that. Of
26:32
course you are. You think, how difficult is that
26:34
problem to have a robot that
26:38
basically can build itself?
26:40
Very, very hard. I think you've mentioned
26:43
this, like, to me
26:45
or somewhere where people
26:47
think it's easy conceptually. And
26:50
then they remember that you're going to have to have a fab.
26:53
Yeah, on board. Of course. So
26:56
3D printer that prints a 3D printer.
26:59
Yeah.
27:00
Yeah, on legs. Why
27:02
is that hard? Well, because it's not, I
27:04
mean a 3D printer is a very simple
27:07
machine, right? Okay, you're going to print chips,
27:09
you're going to have an atomic printer, how are you going to dope
27:12
the silicon? Yeah.
27:14
Right? How are you going to etch the silicon? You're
27:17
going to have to have a
27:19
very interesting kind of fab if you want to have
27:22
a lot of computation on board. But
27:24
you can do, like, structural
27:28
type of robots that are dumb.
27:30
Yeah, but structural type of robots
27:33
aren't going to have the intelligence required to survive
27:35
in any complex environment.
27:36
What about, like, ants type of systems? We
27:39
have, like, trillions of them. I don't
27:41
think this works. I mean, again, like, ants
27:43
at their very core are made up of cells
27:45
that are capable of individually reproducing.
27:48
They're doing quite a lot of computation
27:50
that we're taking for granted. It's not even just
27:52
the computation. It's that reproduction is so
27:55
inherent. Okay, so, like, there's two stacks of life in
27:57
the world. There's the biological
27:59
stack and the silicon stack.
27:59
The biological stack
28:02
starts with reproduction. Reproduction
28:05
is at the absolute core. The first proto
28:08
RNA organisms were capable of reproducing. The
28:11
silicon stack, despite as far
28:13
as it's come, is nowhere near
28:15
being able to reproduce. Yeah.
28:18
So the fab movement, digital
28:23
fabrication,
28:25
fabrication in the full range of what that means is
28:28
still in the early stages. Yeah.
28:30
You're interested in this world. Even
28:32
if you did put a fab on the machine, let's say, okay, we can build
28:35
fabs. We know how to do that as humanity. We
28:37
can probably put all the precursors that build all the machines
28:40
and the fabs also in the machine. So first off, this machine is
28:42
going to be absolutely massive.
28:44
I mean, we almost have a... Think
28:47
of the size of the thing required to reproduce
28:49
a machine today.
28:52
Is our civilization capable of reproduction?
28:55
Can
28:56
we reproduce our civilization on Mars?
29:00
If we were to construct a machine that is made up of humans, like
29:02
a company, it can reproduce
29:04
itself. Yeah. I don't know. It
29:08
feels like 115 people.
29:12
I think it's so much harder than that. Let's
29:16
see. I
29:18
believe that Twitter can be run by 50 people.
29:22
I think that this is going to take most
29:24
of... It's just most
29:27
of society. We live in one globalized
29:29
world. No, but you're not interested in running Twitter.
29:32
You're interested in seeding.
29:33
You want to seed
29:36
a civilization then because humans
29:38
can have sex. Yeah,
29:40
okay. So you're talking about the humans reproducing
29:42
and basically, what's the smallest self-sustaining colony
29:44
of humans? Yeah. Yeah, okay, fine. Over
29:48
time, they will. I think you're being...
29:51
We have to expand our conception of time
29:53
here. Come back to the original. Time
29:56
scale. I mean, over across...
29:59
maybe a hundred generations, we're back to making
30:02
chips. No? If you seed
30:04
the colony correctly.
30:06
Maybe. Or maybe they'll watch our
30:08
colony die out over here and be like, we're
30:10
not making chips, don't make chips. No, but you
30:12
have to seed that colony correctly. Whatever you
30:14
do, don't make chips. Chips are what
30:17
led to their downfall.
30:20
Well, that is the thing that humans do. They
30:22
come up, they construct a devil, a
30:24
good thing and a bad thing, and they really stick by that,
30:27
and then they murder each other over that. There's always
30:29
one asshole in the room who murders everybody. And
30:33
he usually makes tattoos and nice branding. Do
30:35
you need that asshole? That's the question, right?
30:38
Humanity works really hard today to get rid of that asshole,
30:40
but I think they might be important. Yeah,
30:43
this whole freedom of speech thing, it's
30:45
the freedom of being an asshole seems kind of important.
30:47
That's right.
30:49
Man, this thing, this fab, this human
30:51
fab that we've constructed, this human
30:53
civilization, is pretty interesting. And now
30:56
it's
30:56
building artificial copies
30:59
of itself, or artificial copies of
31:01
various aspects of itself
31:04
that seem interesting, like intelligence. And
31:07
I wonder where that goes. I
31:10
like to think it's just like another stack for life. Like
31:12
we have the biostack life, like we're a biostack life,
31:14
and then the silicon stack life. But it seems
31:16
like the ceiling, or there might
31:18
not be a ceiling, or at least the ceiling is much higher
31:21
for the silicon stack. Oh,
31:23
no, we don't know what the ceiling is for the biostack
31:26
either. The biostack just
31:28
seemed to move slower. You have
31:31
Moore's law, which is not dead
31:33
despite many proclamations. In
31:35
the biostack or the silicon stack? In the silicon stack. And
31:38
you don't have anything like this in the biostack. So I have
31:40
a meme that I posted, I tried to make a
31:42
meme, it didn't work too well. But I posted
31:44
a picture of Ronald Reagan and Joe
31:46
Biden, and you look, this is 1980 and this is 2020. And
31:50
these two humans are basically like the same, right?
31:52
There's been no change in humans in the
31:55
last 40 years.
32:00
And then I posted a computer from 1980 and a computer
32:02
from 2020. Wow. Yeah,
32:07
with the early stages, right? Which
32:09
is why you said when you said the fab,
32:11
the size of the fab required to make another
32:13
fab is like
32:16
very large right now. Oh,
32:18
yeah. But computers were very large 80 years
32:23
ago and they got pretty
32:26
tiny. People
32:29
are starting to want to wear them on their face in
32:33
order to escape reality. That's
32:36
the thing in order to be live inside
32:38
the computer. Yeah. Put a
32:40
screen right here. I don't have to see the rest
32:42
of you assholes. I've been ready for a long
32:44
time. You like virtual reality? I love
32:46
it. Do you want to live
32:49
there? Yeah. Yeah.
32:52
Part of me does too. How far away are
32:54
we, do you think? Anything
32:57
from what you can buy today far? Very
33:00
far. I got to tell you that
33:02
I had the experience of
33:06
Meta's Kodak avatar where
33:09
it's an ultra high resolution
33:12
scan. It looked
33:15
real. I mean, the
33:17
headsets just are not quite at like eye resolution
33:19
yet. I haven't put on any
33:22
headset where I'm like, oh, this could be
33:24
the real world. Whereas when
33:26
I put good headphones on, audio
33:27
is there. We can
33:30
reproduce audio that I'm like, I'm actually in a jungle right
33:32
now. If I close my eyes, I can't tell I'm not. Yeah.
33:36
But then there's also smell and all that kind of stuff. Sure.
33:39
I don't know. The power
33:41
of imagination or the power of the
33:43
mechanism in the human mind that fills the gaps
33:46
that kind of reaches and wants to make
33:48
the thing you see in the virtual world real
33:51
to you, I
33:53
believe in that power. Or humans
33:55
want to believe. Yeah.
33:58
What if you're lonely? What if you're sad? What
34:00
if you're really struggling in life and
34:02
here's a world where you don't have to struggle
34:05
anymore Humans want to believe so much
34:07
that people think the large language models are conscious.
34:10
That's how much humans want to believe
34:12
Strong words. He's throwing
34:14
left and right hooks. Why do you
34:17
think large language models are not conscious? I
34:19
don't think I'm conscious. Oh,
34:21
so what is consciousness then George Hans?
34:24
It's like
34:25
what it seems to mean to people. It's just like
34:27
a word that atheists use for souls Sure,
34:31
but that doesn't mean soul is not an interesting word
34:34
If consciousness is a spectrum. I'm definitely way
34:37
more conscious than the large language models are
34:39
I
34:41
Think the large language models are less conscious than
34:43
a chicken When is
34:45
the last time you see a chicken? In
34:48
Miami like a couple months ago. How
34:52
no like a living chicken living chickens walking
34:54
around Miami It's crazy like on the street.
34:56
Yeah, like a chicken chicken All
35:02
right, I was trying to call you all like
35:04
a good journalist and I I got shut
35:06
down Okay,
35:09
but
35:10
you don't think
35:12
much about this kind
35:15
of Subjective
35:18
feeling that it feels like something
35:21
to exist and then as an observer
35:25
you can Have a sense
35:27
that an entity is not only
35:29
intelligent but has a kind of
35:33
Subjective experience of its reality
35:35
like a self-awareness That is capable
35:38
of like suffering of hurting of being excited
35:40
by the environment in a way. That's not
35:42
merely
35:44
Kind of an artificial response but a deeply
35:47
felt one Humans want to believe so
35:50
much that if I took a rock and a sharpie
35:52
and drew a sad face on the rock They'd think the
35:54
rock is sad
35:55
Yeah,
35:58
and you're saying when we look in the mirror we we
36:00
apply the same smiley face with rock.
36:02
Pretty much, yeah. Isn't that weird though?
36:05
You're not conscious?
36:09
But you do believe in consciousness. Really?
36:12
It's just, it's unclear. Okay, so to you it's like
36:14
a little, like a symptom of the bigger
36:16
thing that's not that important. Yeah, I mean
36:18
it's interesting that like human systems
36:21
seem to claim that they're conscious. And I guess
36:23
it kind of like says something and they straight up like,
36:25
okay, what do people mean? Even if you don't believe
36:27
in consciousness, what do people mean when they say consciousness?
36:30
And there's definitely like
36:31
meanings to it. What's your favorite
36:33
thing to eat? Pizza.
36:37
Cheese pizza, what are the toppings? I like cheese pizza.
36:40
Don't say pineapple. No, I don't like pineapple. Okay,
36:42
pepperoni pizza. Has they put any ham on it? Oh, that's real
36:44
bad. What's the best pizza? What
36:47
are we talking about here? Like, do you like cheap crappy pizza?
36:50
Chicago deep dish cheese pizza. Oh,
36:52
that's my favorite. There you go, you bite into a deep
36:54
dish, a cargo deep dish pizza,
36:56
and it feels like you were starving. You haven't
36:59
eaten for 24 hours. You just bite
37:01
in and you're hanging out with somebody that
37:03
matters a lot to you and you're there with the pizza. Sounds
37:05
real nice, huh? Yeah, all right. It
37:08
feels like something. I'm George
37:10
motherfucking hot eating a fucking
37:13
Chicago deep dish pizza. There's
37:15
just a full peak
37:17
living experience of
37:20
being human, the top of the human condition.
37:23
Sure. It feels like something
37:25
to experience that. Why
37:28
does it feel like something? That's consciousness,
37:31
isn't it? If that's the word
37:33
you want to use to describe it, sure. I'm not going to deny
37:35
that that feeling exists. I'm not going to deny that I experienced
37:38
that feeling. When, I
37:40
guess what I kind of take issue to is that
37:42
there's some like, like how
37:44
does it feel to be a web server? Do 404s hurt?
37:49
Not yet. How would you know what suffering
37:51
looked like? Sure, you can recognize a suffering
37:53
dog because we're the same stack as the dog. All
37:56
the bio stack stuff, especially mammals,
37:59
it's really easy. You can... Game
38:02
recognizes game. Yeah. Versus
38:04
the silicon stack stuff. It's like, you
38:06
have no idea.
38:08
You have... Wow, the little thing
38:10
has learned to mimic.
38:15
But then I realized that that's all we are too.
38:18
Oh look, the little thing has learned to mimic. Yeah.
38:21
I guess, yeah, 404 could be suffering, but it's so far from our kind
38:23
of living.
38:27
Living
38:30
organism, our kind of stack. But
38:32
it feels like AI can start
38:35
maybe mimicking the biological
38:37
stack better, better, better. Because it's trained. Retrained
38:39
it, yeah. And so, in
38:41
that, maybe that's the definition of consciousness. Is
38:44
the bio stack consciousness? The definition
38:46
of consciousness is how close something looks to human.
38:48
Sure, I'll give you that one.
38:50
No, how close something
38:52
is to the human experience. Sure.
38:55
It's a very anthropocentric
38:58
definition, but... That's all we got. Sure.
39:01
No, and I don't mean to like... I think there's
39:03
a lot of value in it. Look, I just started my second company.
39:05
My third company will be AI Girlfriends.
39:08
No, like I mean it. I want to find out what your fourth company
39:11
is after that. Oh wow. Because I think once
39:13
you have AI Girlfriends, it's... Oh
39:17
boy. Does it get interesting?
39:20
Well, maybe let's go there. I mean, the relationships
39:22
with AI, that's creating human-like
39:24
organisms, right?
39:27
And part of being human is being conscious, is
39:30
having the capacity to suffer, having the capacity
39:32
to experience this life richly in
39:34
such a way that you can empathize...
39:36
The AI system can empathize with you
39:38
and you can empathize with it. Or you can
39:40
project your anthropomorphic
39:44
sense of what the other entity is experiencing.
39:48
And an AI model would need
39:50
to create that experience inside your mind. And
39:53
it doesn't seem that difficult. Yeah, but okay.
39:55
So here's where it actually gets totally different, right?
39:59
When you interact... with another human, you can
40:01
make some assumptions.
40:04
When you interact with these models, you can't. You
40:06
can make some assumptions that that other human experiences
40:09
suffering and pleasure in a pretty
40:11
similar way to you do. The golden rule applies.
40:15
With an AI model, this isn't really true.
40:18
These large language models are good at fooling people
40:20
because they were trained on a whole bunch
40:23
of human data and told to mimic it.
40:25
But if the AI system says,
40:28
hi, my name is Samantha,
40:31
it has a backstory. I went to college
40:33
here and there. Maybe you'll integrate
40:36
this in the AI system. I made some chatbots.
40:38
I gave them backstories. It was lots of fun. I
40:40
was so happy when Llama came out. Yeah. We'll
40:43
talk about Llama. We'll talk about all that. But like,
40:45
you know, the rock with the smiley face. It
40:49
seems pretty natural for you to anthropomorphize
40:51
that thing and then start dating it. Before
40:55
you know it, you're married and
40:58
have kids. With a rock? With
41:00
a rock. And there's pictures on Instagram
41:02
with you and a rock and a smiley face. To
41:04
be fair, like, you know, something that people generally look
41:06
for in the look of someone to date is intelligence
41:08
in some
41:10
form. And the rock doesn't really have
41:12
intelligence. Only a pretty desperate person would date a rock.
41:16
I think we're all desperate deep down. Oh,
41:18
not rock level desperate. All
41:20
right. Not
41:23
rock level desperate,
41:26
but AI level
41:28
desperate. I don't know. I think all
41:30
of us have a deep loneliness. It just feels
41:32
like the language models are there. Oh,
41:35
I agree. And you know what? I won't even say this so cynically.
41:37
I will actually say this in a way that like, I want AI
41:39
friends. I do. Yeah. Like,
41:42
I would love to. You know, again, the
41:44
language models now are still a little like,
41:48
people are impressed with these GPT things.
41:50
And I look at like, or like, or
41:53
the copilot, the coding one. And
41:55
I'm like, okay, this is like junior engineer
41:57
level. And these people are like fiverr level.
41:59
artists and copywriters.
42:02
Like, okay, great, we got like
42:04
Fiverr and like junior engineers, okay,
42:06
cool. Like, and this is just the start
42:09
and it will get better, right? Like
42:11
I can't wait to have AI friends who
42:13
are more intelligent than I am.
42:15
So Fiverr is just a temporary, it's not the
42:17
ceiling. No, definitely not. Is
42:21
it countless cheating
42:23
when you're talking to an AI model, emotional
42:26
cheating? That's
42:30
up to you and your human partner to
42:32
define. Oh, you have to, all right. Yeah,
42:35
you have to have that conversation, I guess.
42:37
All right, I mean, integrate that
42:40
with porn and all this stuff. No,
42:42
I mean, it's similar kind of to porn. Yeah. Yeah. Right,
42:45
I think people in relationships have different views on that.
42:47
Yeah, but most people don't
42:50
have like serious
42:54
open conversations about all the different
42:57
aspects of what's cool and what's not. And
43:00
it feels like AI is a really weird conversation
43:02
to have. The
43:04
porn one is a good branching off point. Like
43:06
these things, you know, one of my scenarios that I put in my chatbot
43:09
is I go, you know, a
43:12
nice girl named Lexi, she's 20, she just moved
43:14
out to LA, she wanted to be an actress, but she
43:16
started doing OnlyFans instead and you're on a date with her,
43:18
enjoy. Oh,
43:22
man. Yeah,
43:24
and so is that if you're actually dating somebody
43:26
in real life, is that cheating? I
43:29
feel like it gets a little weird. Sure. It gets
43:31
real weird. It's like, what are you allowed to
43:33
say to an AI bot? Imagine having
43:35
that conversation with a significant other. I mean,
43:37
these are all things for people to define in their relationships.
43:40
What it means to be human is just gonna start to
43:42
get weird. Especially online.
43:44
Like, how do you know? Like, there'll
43:46
be moments when you'll have what you
43:48
think is a real human you interacted
43:50
with on Twitter for years and you realize it's not.
43:53
I spread, I love this meme,
43:56
heaven banning.
43:57
Tell you about shadow banning. Yeah. Shadow
44:00
banning, okay, you post, no one can see it. Heaven
44:02
banning, you post, no one can see
44:04
it, but a whole lot of AIs are spot
44:06
up to interact with you. Well,
44:10
maybe that's what the way human civilization ends
44:12
is all of us are heaven banned. There's
44:15
a great, it's called My Little Pony
44:17
Friendship is Optimal.
44:18
It's a sci-fi story that explores
44:21
this idea. Friendship is optimal. Friendship
44:24
is optimal. Yeah, I'd like to have some, at least
44:26
on the intellectual realm, from AI friends
44:29
that argue with me.
44:30
But the romantic realm
44:33
is weird, definitely weird.
44:38
But not out of the realm of the
44:41
kind of weirdness that human civilization is capable
44:44
of, I think. I want
44:46
it. Look, I want it. If no one else wants
44:48
it, I want it. Yeah, I think a lot of people probably
44:50
want it. There's a deep loneliness.
44:53
And I'll feel their loneliness
44:56
and, you know, it's just, we'll only advertise to you
44:58
some of the time. Yeah, maybe the conceptions
45:00
of monogamy change too. Like, I grew up
45:02
in a time, like, I value monogamy, but maybe
45:04
that's a silly notion when you have
45:07
arbitrary number of AI systems. Mm,
45:10
this interesting path from rationality
45:13
to polyamory, that doesn't make sense
45:15
for me. For you. But you're just
45:17
a biological organism who was born before,
45:20
like, the internet really took off.
45:23
The crazy thing is, like,
45:25
culture is whatever we define it as.
45:28
These things are not, like,
45:31
is a lot problem in moral philosophy, right? There's
45:33
no, like, okay, what is might be that, like,
45:36
computers are capable of mimicking, you
45:39
know, girlfriends perfectly. They pass the girlfriend Turing
45:41
test, right? But that doesn't say anything about
45:43
a lot. That doesn't say anything about how we ought to
45:45
respond to them as a civilization. That doesn't say we ought
45:47
to get rid of monogamy, right? That's a
45:50
completely separate question, really a religious
45:52
one.
45:52
And Turing test, I wonder
45:54
what that looks like. Girlfriend Turing test. Are you
45:57
writing that? Will
45:59
you be the... the Alan Turing of the
46:01
21st century that writes the girlfriend Turing
46:03
test paper? No, I mean, of course, my, hey, girlfriends,
46:06
their goal is to pass the girlfriend Turing test.
46:09
No, but there should be like a paper that kind
46:11
of defines the test. Or,
46:14
I mean, the question is if it's deeply personalized
46:16
or there's a common thing that really gets everybody.
46:21
Yeah, I mean, you know, look, we're a company,
46:23
we don't have to get everybody, we just have to get a large enough clientele
46:26
to stay. I like how you're already thinking
46:28
company. All right, let's, before
46:30
we go to company number three and company number
46:32
four, let's go to company number two. All right.
46:35
TinyCorp,
46:37
possibly one of the greatest names of all
46:39
time for a company. You've
46:41
launched a new company called
46:43
TinyCorp that leads the development of TinyGrad.
46:46
What's the origin story
46:48
of TinyCorp and TinyGrad?
46:50
I started TinyGrad as
46:53
like a toy project, just to teach myself,
46:55
okay, like, what is a convolution? What
46:58
are all these options you can pass to them? What is
47:00
the derivative of a convolution, right? Very similar
47:03
to Carpathi wrote MicroGrad. Very
47:06
similar. And
47:08
then
47:09
I started realizing, I started thinking
47:11
about like
47:12
AI chips. I started thinking about chips that
47:14
run AI and I
47:17
was like, well, okay, this is going to be
47:19
a really big problem. If Nvidia
47:22
becomes a monopoly here, how
47:25
long before Nvidia is nationalized?
47:28
So you, one
47:30
of the reasons that start
47:33
TinyCorp is to challenge Nvidia.
47:36
It's not so much
47:37
to challenge Nvidia. I actually, I
47:39
like Nvidia and it's to make
47:42
sure power
47:45
stays decentralized. Yeah.
47:48
And here's computational
47:50
power. I see you
47:52
Nvidia is kind of locking down the
47:54
computational power of the world.
47:56
If Nvidia becomes
47:58
just like 10X better than everything.
47:59
else, you're giving a big advantage
48:02
to somebody who can secure NVIDIA
48:05
as a resource. Yeah.
48:07
In fact,
48:08
if Jensen watches this podcast, he may want to consider
48:11
this.
48:12
He may want to consider making sure his company is not
48:14
nationalized. Do
48:16
you think that's an actual threat? Oh, yes. No,
48:20
but there's so much, you know, there's
48:23
AMD. So we have NVIDIA and AMD,
48:25
great. All right.
48:27
You don't think there's like a push
48:31
towards like selling, like Google
48:33
selling TPUs or something like this? You
48:35
don't think there's a push for that? Have you seen it? Google
48:37
loves to rent you TPUs. It
48:39
doesn't. You can't buy it at Best Buy?
48:42
No.
48:43
So I started work on a chip. I
48:46
was
48:48
like, okay, what's it going to take to make a chip? And
48:51
my first notions were all completely wrong about why,
48:53
about like how you could improve on GPUs. And
48:56
I will take this. This is from Jim
48:58
Keller on your podcast. And
49:00
this is one of my absolute favorite
49:03
descriptions of computation.
49:05
So there's three kinds of computation paradigms
49:08
that are common in the world today.
49:10
They're CPUs. And
49:12
CPUs can do everything. CPUs can do
49:14
add and multiply.
49:15
They can do load and store, and they can do
49:17
compare and branch. And when I say they can
49:19
do these things, they can do them all fast, right? So
49:22
compare and branch are unique to CPUs. And
49:24
what I mean by they can do them fast is they can do things like
49:27
branch prediction and speculative execution. And they
49:29
spend tons of transistors and they use like super deep
49:31
reorder buffers in order to make these things
49:33
fast. Then you have a simpler computation
49:36
model GPUs. GPUs can't really do
49:38
compare and branch. I mean, they can, but it's horrendously
49:40
slow. But GPUs can do arbitrary
49:42
load and store. GPUs can do things
49:45
like X, D reference Y.
49:47
So they can fetch from arbitrary pieces of memory. They
49:49
can fetch from memory that is defined by the contents of the data.
49:52
The third model of computation
49:54
DSPs and DSPs are just add and
49:56
multiply.
49:57
Like they can do load and store, but only static load.
49:59
stores, only loads and stores that are known before
50:02
the program runs. And you look at neural
50:04
networks today and 95% of neural networks
50:06
are all the DSP paradigm,
50:09
they are just statically scheduled
50:12
ads and multiplies. So
50:14
TinyGuard really took this idea and,
50:17
and I'm still working on it to extend this as
50:19
far as possible. Um, every
50:21
stage of the stack has Turing completeness, right? Python
50:24
has Turing completeness. And then we take Python and
50:26
we go into C plus plus, which is Turing complete. And
50:28
maybe C plus plus calls into some CUDA kernels,
50:30
which are turn complete, the CUDA kernels go through LLVM,
50:33
which is turn complete into PTX, which is turn complete
50:35
to SAS, which is turn complete on a current turn complete processor.
50:37
I want to get Turing completeness out of the stack entirely.
50:40
Because once you get rid of Turing completeness, you can reason about
50:42
things. Rice's theorem and the halting problem
50:45
do not apply to admiral machines.
50:47
Okay. What's
50:50
the power and the value of getting Turing completeness out
50:52
of, out of, are we talking about
50:54
the hardware or the software? Every
50:57
layer of the stack, every layer
50:59
of the stack, removing turn completeness allows
51:01
you to reason about things, right? So
51:03
the reason you need to do branch prediction in a CPU
51:06
and the reason it's prediction and the branch predictors are,
51:08
I think they're like 99% on CPUs. Why
51:10
did they get 1% of them wrong? Well, they
51:12
get 1% wrong because you can't
51:15
know.
51:15
Right. That's the halting problem. It's equivalent
51:18
to the halting problem to say whether a branch is going
51:20
to be taken or not. Um, I can
51:22
show that, but
51:24
the. Admiral
51:26
machine, the neural network runs
51:28
the identical compute every time. The
51:31
only thing that changes is the data. So
51:34
when you realize this, you think about, okay,
51:37
how can we build a computer? How can we build
51:39
a stack that takes maximal advantage of
51:41
this idea? Uh,
51:43
so
51:44
what makes tiny grad different from other
51:46
neural network libraries is it does not have
51:49
a primitive operator even for matrix multiplication.
51:52
And this is every single one. They
51:54
even have primitive operators for things like convolutions.
51:56
So no mat mall. No, Matt. Well,
51:59
here's what a.
51:59
I'm at my list, so I'll use my hands to talk here.
52:02
So if you think about a cube, and I put my two
52:04
matrices that I'm multiplying on two faces of the
52:07
cube, you can
52:09
think about the matrix multiply as, okay,
52:12
the n cubed, I'm going to multiply for
52:14
each one in the cubed, and then I'm going to do a sum, which
52:16
is a reduce up to here to the third
52:18
face of the cube, and that's your multiplied matrix. So
52:21
what a matrix multiply is, is a bunch of shape
52:24
operations, right? A bunch of permutes,
52:26
reshapes, and expands on the two matrices.
52:29
A multiply, n cubed, a
52:31
reduce, n cubed, which gives
52:33
you an n squared matrix. Okay, so
52:36
what is the minimum number of operations that can accomplish
52:38
that if you don't have MatMall
52:40
as a primitive? So TinyGrad has
52:43
about 20, and you can compare TinyGrad's
52:46
op-set or IR to things like XLA
52:49
or PrimTorch. So XLA and
52:51
PrimTorch are ideas where like, okay, Torch
52:54
has like 2000 different kernels,
52:58
PyTorch 2.0 introduced PrimTorch,
53:00
which has only 250.
53:01
TinyGrad has order of magnitude 25.
53:05
It's 10X less than
53:07
XLA or PrimTorch. And
53:09
you can think about it as kind of like RISC versus CISC,
53:12
right?
53:13
These other things are CISC-like
53:15
systems. TinyGrad is RISC.
53:18
And RISC-1. RISC architecture
53:21
is going to change everything. 1995 hackers. Wait,
53:25
really? That's an actual thing? Angelina
53:27
Jolie delivers the line, RISC architecture
53:29
is going to change everything in 1995. Wow. And
53:32
here we are with ARM in the phones and
53:34
ARM everywhere. Wow, I
53:36
love it when movies actually have real things in
53:38
them. Right. Okay, interesting.
53:41
So you're thinking of this as the RISC
53:44
architecture of ML
53:45
Stack. 25, huh?
53:49
What can you go through the four
53:51
op types?
53:56
Okay, so you have UnaryOps, which
53:59
take in... a Tensor
54:02
and return a tensor of the same size and do some
54:04
unary opt to it X log Reciprocal
54:08
sign right they take in one and their point
54:10
wise Really you
54:13
yeah really you almost all activation
54:15
functions are unary ops Some
54:17
combinations of unary ops together
54:19
is still a unary op
54:22
Then you have binary ops binary ops are like
54:24
point wise addition multiplication division compare
54:28
It takes in two tensors of equal size and
54:31
outputs one tensor Then
54:33
you have reduce ops Reduce
54:36
ops will like take a three-dimensional tensor and
54:38
turn it into a two-dimensional tensor Or
54:40
three-dimensional tensor turn it to zero dimensional tensor
54:42
think like a sum or max
54:44
are really the common ones there And
54:47
then the fourth type is movement ops and
54:50
movement ops are different from the other types because they don't actually
54:52
require computation They require different ways to look
54:54
at memory So that includes reshapes
54:57
permutes Expans flips
55:00
does the main ones probably so with that you have enough
55:02
to make a map mall and convolutions
55:05
and every convolution you can Imagine dilated
55:07
convolutions strided convolutions transposed
55:10
convolutions
55:12
You're right on github about laziness
55:16
showing a map mall Matrix
55:18
multiplication see how despite the style
55:20
is fused into one kernel with
55:23
the power of laziness Can you elaborate
55:25
on this power of laziness sure so if
55:27
you type in Pytorch a times
55:30
B plus C?
55:32
What this is going to do is
55:34
it's going to first multiply add and
55:36
be a and B and store that result
55:38
into memory
55:39
and then it is going to add C by reading
55:42
that result from memory reading C for memory
55:44
and writing that out to memory There
55:47
is way more loads and stores to memory than
55:49
you need there If you don't actually do
55:52
a times B as soon as you see it
55:54
if you wait
55:55
Until the user actually realizes that
55:58
tensor until the laziness actually resolved You
56:00
can fuse that plus C. This is like,
56:03
it's the same way Haskell works. So
56:05
what's the process of porting a model
56:08
into TinyGrad? So TinyGrad's
56:10
front end looks very similar to PyTorch. I
56:13
probably could make a perfect,
56:15
or pretty close to perfect interop layer if I
56:17
really wanted to. I think that there's some things that
56:19
are nicer about TinyGrad syntax than PyTorch, but
56:22
the front end looks very Torch-like. You can also
56:24
load in Onyx models. We have
56:27
more Onyx tests passing than Core ML. Core
56:30
ML. Okay, so- We'll pass Onyx
56:32
runtime soon. What about the
56:34
developer experience with TinyGrad? What
56:38
it feels like versus PyTorch?
56:40
By the way, I
56:43
really like PyTorch. I think that it's actually a very
56:45
good piece of software. I think that they've
56:47
made a few different trade-offs, and
56:49
these different trade-offs are
56:52
where, TinyGrad takes
56:54
a different path. One of the biggest differences
56:56
is it's really easy to see the kernels
56:59
that are actually being sent to the GPU.
57:00
If you run PyTorch on
57:02
the GPU,
57:04
you do some operation, and you don't know what
57:06
kernels ran. You don't know how many kernels ran. You
57:08
don't know how many flops were used. You don't know how much
57:10
memory accesses were used. TinyGrad
57:12
type debug equals two, and it
57:14
will show you in this beautiful style every
57:17
kernel that's run.
57:19
How many flops
57:21
and how many bytes. So
57:24
can you just linger
57:26
on what problem
57:28
TinyGrad solves? TinyGrad
57:30
solves the problem of porting new ML
57:32
accelerators quickly. One
57:34
of the reasons, tons
57:37
of these companies now, I think Sequoia
57:40
marked Graphcore to zero.
57:42
Seribis, TenzTorrent,
57:44
Grok, all of
57:46
these ML accelerator companies,
57:49
they built chips. The chips were good. The
57:51
software was terrible. And
57:54
part of the reason is because I think the same problem is happening
57:56
with Dojo. It's really, really
57:58
hard to write a PyTorch port.
57:59
because you have to write 250 kernels
58:03
and you have to tune them all for performance. What
58:06
does Jim Keller think about Tanya Grad?
58:08
You guys hung on
58:10
quite a bit, so he's, you know, he
58:13
was involved with Tencentorrent. What's
58:15
his praise and what's his criticism
58:18
of what you're doing with your life?
58:20
Look, my
58:23
prediction for Tencentorrent is that they're gonna pivot to making
58:25
RISC-V chips. CPUs.
58:29
CPUs. Why?
58:33
Because AI accelerators are
58:35
a software problem, not really a hardware problem. Oh,
58:38
interesting, so you don't think, you
58:40
think the diversity of AI accelerators
58:43
in the hardware space is not going to be
58:45
a thing that exists long-term.
58:47
I think what's gonna happen is if
58:49
I can, okay,
58:51
if you're trying to make an AI accelerator,
58:54
you better have the capability
58:56
of writing a torch-level
58:58
performance stack on NVIDIA GPUs.
59:01
If you can't write a torch stack on NVIDIA
59:03
GPUs, and I mean all the way, I mean down to the driver,
59:05
there's no way you're gonna be able to write it on your chip because
59:08
your chip's worse than an NVIDIA GPU. The first
59:10
version of the chip you tape out, it's definitely worse. Oh,
59:12
and you're saying writing that stack is really tough. Yes,
59:15
and not only that, actually, the chip that you tape out, almost
59:17
always because you're trying to get advantage over NVIDIA, you're
59:19
specializing the hardware more. It's
59:21
always harder to write software for more specialized
59:24
hardware. Like a GPU's pretty generic,
59:26
and if you can't write an NVIDIA stack, there's
59:28
no way you can write a stack for your chip. So
59:31
my approach with TinyGrad is first, write
59:33
a performant NVIDIA stack. We're targeting
59:36
AMD. So
59:38
you did say a few to NVIDIA a little bit, with
59:41
love. With love. Yeah, but so what- It's
59:43
like the Yankees, you know? I'm a Mets fan. Oh,
59:46
you're a Mets fan. A risk
59:48
fan and a Mets fan. What's the hope that AMD
59:51
has? You did a build
59:53
with AMD recently that I saw. That
59:56
was the PTSD. There,
59:59
the 79,
59:59
700 XTX compared to the
1:00:02
RTX 4090 or 4080. Well,
1:00:04
let's start with the fact that the 7900 XTX kernel
1:00:07
drivers don't work. And if you run demo
1:00:09
apps in loops, it panics the kernel.
1:00:11
Okay, so this is a softer issue.
1:00:15
Lisa Sue responded to my email.
1:00:17
Oh. I reached out, I was like,
1:00:19
this is, you know, really?
1:00:22
Like, I understand if your seven
1:00:25
by seven transposed Winograd comm
1:00:27
is slower than Nvidia's, but literally when
1:00:29
I run demo apps in a loop, the
1:00:31
kernel panics.
1:00:33
So just adding that loop.
1:00:36
Yeah, I just literally took their demo apps and wrote like, while
1:00:38
true, semi-colon do the app, semi-colon
1:00:41
done in a bunch of screens. Right,
1:00:43
this is like the most primitive fuzz testing.
1:00:46
Why do you think that is? They're just
1:00:48
not seeing a market in
1:00:51
machine learning? They're changing,
1:00:53
they're trying to change. They're trying to change.
1:00:55
And I had a pretty positive interaction with them this week.
1:00:57
Last week I went on YouTube, I was just like, that's
1:00:59
it. I give up on AMD. Like, this is their
1:01:02
driver,
1:01:02
doesn't even, I'm not gonna, I'll
1:01:05
go with Intel GPUs. Intel GPUs have
1:01:07
better drivers. So
1:01:10
you're kind of spearheading the
1:01:13
diversification of GPUs.
1:01:16
Yeah, and I'd like to extend that diversification
1:01:18
to everything. I'd like to diversify
1:01:20
the, right, the more,
1:01:25
my central thesis about the world is,
1:01:28
there's things that centralize power and they're bad. And
1:01:30
there's things that decentralize power and they're good.
1:01:33
Everything I can do to help decentralize power,
1:01:35
I'd like to do.
1:01:38
So you're really worried about the centralization of Nvidia, that's interesting.
1:01:41
And you don't have a fundamental hope for the
1:01:44
proliferation of ASICs, except
1:01:46
in the cloud.
1:01:49
I'd like to help them with software. No, actually, there's
1:01:51
only, the only ASIC that is remotely successful
1:01:54
is Google's TPU. And the only
1:01:56
reason that's successful is because Google wrote
1:01:58
a
1:01:59
machine learning framework.
1:02:00
I think that you have to write a competitive
1:02:02
machine learning framework in order to be able
1:02:04
to build an ASIC.
1:02:07
You think Meta with PyTorch builds
1:02:09
a competitor? I hope so. They
1:02:12
have one. They have an internal one. Internal.
1:02:14
I mean, public facing with a nice cloud
1:02:16
interface and so on. I don't want
1:02:18
a cloud. You don't like cloud. I don't
1:02:20
like cloud. What do you think is the fundamental
1:02:22
limitation of cloud? Fundamental limitation
1:02:25
of cloud is who owns the off switch.
1:02:27
So it's a power to the people. Yeah.
1:02:30
And you don't like the man to have all the
1:02:32
power. Exactly. All right.
1:02:35
And right now, the only way to do that is with AMD
1:02:37
GPUs if you want performance and
1:02:39
stability. Interesting.
1:02:43
But it's a costly investment emotionally
1:02:45
to go with AMDs. Well,
1:02:48
let me add sort of on a tangent to ask you, what did,
1:02:52
you've built quite a few PCs. What's your advice
1:02:54
on how to build a good custom PC for
1:02:57
let's say for the different applications they use for
1:02:59
gaming, for machine learning? Well, you
1:03:01
shouldn't build one. You should buy a box from the tiny Corp.
1:03:04
I heard rumors, whispers
1:03:08
about this box in the tiny Corp.
1:03:10
What's this thing look like? What is it?
1:03:12
What is it called? It's called the tiny box. Tiny
1:03:14
box? It's $15,000. And
1:03:18
it's almost a pit of flop of compute. It's
1:03:21
over a hundred gigabytes of GPU RAM. It's
1:03:23
over five terabytes per second of
1:03:26
GPU memory bandwidth.
1:03:29
I'm gonna put like four NVMe's in
1:03:31
RAID. You're gonna get
1:03:34
like 20, 30 gigabytes per second of drive read bandwidth.
1:03:38
I'm gonna build like the best
1:03:40
deep learning box that I can that
1:03:42
plugs into one wall outlet.
1:03:45
Okay. Can you go through those specs again a little
1:03:47
bit from memory? Yeah,
1:03:49
so it's almost a pit of flop of compute. So in
1:03:51
D and tell? Today I'm
1:03:53
leaning toward AMD, but
1:03:56
we're pretty agnostic to the type of compute.
1:03:59
The main limiting spec is a 120 volt 15
1:04:02
amp circuit.
1:04:06
Okay. Well, I mean it, because in order to like,
1:04:09
there's a plug over there, right?
1:04:12
You have to be able to plug it in.
1:04:14
We're also gonna sell the tiny rack, which
1:04:17
like, what's the most power you can
1:04:19
get into your house without arousing suspicion? And
1:04:22
one of the answers is an electric car
1:04:24
charger.
1:04:25
Wait, where does the rack go? Your
1:04:27
garage. Interesting. The
1:04:30
car charger.
1:04:31
A wall outlet is about 1500 watts. A
1:04:34
car charger is about 10,000 watts.
1:04:36
What is the most amount
1:04:38
of power you can get your hands on
1:04:40
without arousing suspicion? That's right.
1:04:42
George Haas. Okay. So
1:04:46
the tiny box and you said NVMe's and RAID.
1:04:49
I forget what you said about memory, all that kind of
1:04:51
stuff. Okay. What about with
1:04:53
GPUs? Again, probably 7900 XTX's,
1:04:58
but maybe 3090's, maybe A770's.
1:05:01
Those are intense. You're flexible
1:05:03
or still exploring? I'm still
1:05:05
exploring. I wanna deliver
1:05:07
a really good experience to people. And
1:05:11
yeah, what GPUs I end up going with. Again, I'm
1:05:13
leaning toward AMD.
1:05:14
We'll see. You know, in my
1:05:16
email, what I said to AMD is like,
1:05:19
just dumping the code on GitHub is not open
1:05:21
source.
1:05:22
Open source is a culture. Open
1:05:24
source means that your issues are not all
1:05:27
one year old stale issues. Open
1:05:29
source means developing
1:05:31
in public. And if you guys can commit
1:05:33
to that, I see a real future for
1:05:35
AMD as a competitor to Nvidia. Well,
1:05:39
I'd love to get a tiny box at MIT.
1:05:41
So whenever it's ready, let's
1:05:43
do it. We're taking pre-orders. I took this from Elon.
1:05:46
I'm like, all right, $100 fully refundable
1:05:48
pre-orders. Is it gonna be like the Cybertruck?
1:05:50
It's gonna take a few years or? No, I'll
1:05:52
try to do it faster. It's a lot simpler. It's a lot simpler
1:05:54
than the truck. Well, there's complexities
1:05:57
not to just the putting
1:05:59
the thing together.
1:05:59
about shipping and all this kind of stuff. The thing
1:06:02
that I want to deliver to people out of the box is
1:06:04
being able to run 65 billion parameter
1:06:06
llama in FP16
1:06:08
in real time, in a good, like 10 tokens
1:06:10
per second or five tokens per second or something. Just,
1:06:12
it works. Yep, just works. Llama's
1:06:15
running or something
1:06:17
like llama. Experience, yeah,
1:06:19
or I think Falcon is the new one, experience
1:06:22
a chat with the largest language model that
1:06:24
you can have in your house.
1:06:26
Yeah, from a wall plug. From
1:06:28
a wall plug, yeah. Actually, for inference,
1:06:31
it's not like even more power would help you get more.
1:06:34
Even more power wouldn't get you more. Well,
1:06:37
no, there's just the biggest, the biggest model released is 65
1:06:39
billion parameter llama as far as I know.
1:06:42
So it sounds like Tiny Box will naturally pivot
1:06:44
towards company number three, because you could just
1:06:46
get the girlfriend and,
1:06:50
or boyfriend. That
1:06:52
one's harder, actually. The boyfriend is harder? Boyfriend's
1:06:54
harder, yeah. I think that's a
1:06:56
very biased statement. I think
1:06:58
a lot of people would just say, why
1:07:01
is it harder to replace
1:07:03
a boyfriend than a other girlfriend
1:07:05
with the artificial llm? Because women
1:07:07
are attracted to status and power and men are
1:07:09
attracted to youth and beauty. No,
1:07:13
I mean, that's what I mean. Both are
1:07:15
a mimicable easy through the language model.
1:07:17
No, no machines do not
1:07:19
have any status or real power.
1:07:21
I don't know, I think you both, well,
1:07:24
first of all, you're using language mostly to
1:07:29
communicate youth
1:07:31
and beauty and power and status. But
1:07:33
status fundamentally is a zero sum game,
1:07:35
whereas youth and beauty are not.
1:07:37
No, I think status is a narrative you can construct.
1:07:40
I don't think status is real. I
1:07:44
don't know. I just think that that's why it's harder.
1:07:47
You know, yeah, maybe it is my biases. I
1:07:49
think status is way easier to fake. I
1:07:51
also think that, you know, men are probably
1:07:53
more desperate and more likely to buy my products, so
1:07:55
maybe they're a better target market. Desperation
1:07:58
is interesting, easier to fool.
1:07:59
Cool. Yeah. I could
1:08:02
see that. Yeah, look, I mean, look, I know you can look at porn
1:08:04
viewership numbers, right?
1:08:05
A lot more men watch porn than women.
1:08:07
You can ask why that is. Wow,
1:08:09
there's a lot of questions and answers
1:08:12
you can get there. Anyway,
1:08:15
with the TinyBox, how
1:08:17
many GPUs in TinyBox? Six.
1:08:20
Ha ha ha ha ha ha ha ha. Oh
1:08:24
man. And I'll tell you why it's six. Yeah. So
1:08:27
AMD Epic processors have 128 lanes of PCIe.
1:08:31
I want to leave enough lanes
1:08:34
for
1:08:35
some drives.
1:08:38
And I want to leave enough lanes for some networking.
1:08:41
How do you do cooling for something like this? Ah,
1:08:44
that's one of the big challenges. Not only
1:08:46
do I want the cooling to be good, I want it to be quiet.
1:08:48
I want the TinyBox to be able to sit comfortably
1:08:50
in your room, right? This is really going towards
1:08:53
the girlfriend thing. So,
1:08:55
because you want to run the LOM. I'll
1:08:57
give a more, I mean, I can talk about how it relates
1:08:59
to company number one.
1:09:01
Common AI. Yeah. Well,
1:09:05
but yes, quiet. Oh, quiet because you may
1:09:07
be potentially want to run it in a car. No, no
1:09:09
quiet because you want to put this thing in your house and you
1:09:11
want it to coexist with you. If it's screaming at 60 dB,
1:09:14
you don't want that in your house, you'll kick it out. 60 dB, yeah. I
1:09:17
want like 40, 45. So how do you make the cooling
1:09:20
quiet? That's an interesting problem in itself.
1:09:22
A key trick is to actually make it big. Ironically,
1:09:25
it's called the TinyBox. But if I can make
1:09:27
it big, a lot of that noise is generated
1:09:29
because of high pressure. If
1:09:31
you look at like a 1U server, a
1:09:33
1U server has these super high pressure fans that
1:09:35
are like super deep and they're like Genesis. Versus
1:09:38
if you have something that's big,
1:09:40
well, I can use a big thing. You know, they call
1:09:42
them big ass fans. Those ones that are like huge on
1:09:44
the ceiling and they're completely
1:09:46
silent. So TinyBox
1:09:48
will be big. I
1:09:52
do not want it to be large according to UPS.
1:09:54
I want it to be shippable as a normal package, but that's my
1:09:57
constraint there.
1:09:58
Interesting. The fans stuff,
1:10:01
can't it be assembled on location or no? No.
1:10:04
No, it has to be, wow. You're... Look,
1:10:07
I wanna give you a great out of the box experience. I want you to lift
1:10:09
this thing out. I want it to be like the Mac, you know?
1:10:12
Tiny box. The Apple experience.
1:10:14
Yeah. I love it. Okay,
1:10:17
and so Tiny Box would run
1:10:20
Tiny Grad. Like what
1:10:23
do you envision this whole thing to look like? We're
1:10:25
talking about like
1:10:26
Linux with the full...
1:10:30
Software engineering environment.
1:10:33
And it's just not PyTorch but Tiny Grad.
1:10:36
Yeah, we did a poll of people want you Bunto or
1:10:38
Arch. We're gonna stick with you Bunto. Ooh,
1:10:40
interesting. What's your favorite flavor
1:10:43
of Linux? You Bunto. Bunto. I
1:10:45
like you Bunto Mate, however you pronounce that,
1:10:47
meat. So
1:10:49
how do you, you've gotten llama into
1:10:52
Tiny Grad. You've gotten stable diffusion into
1:10:54
Tiny Grad. What was that like? Can you comment on like,
1:10:59
what are these models? What's interesting about porting
1:11:01
them? So what's, yeah,
1:11:03
like what are the challenges? What's
1:11:05
naturally, what's easy, all that kind of stuff. There's a
1:11:08
really simple way to get these models into Tiny
1:11:10
Grad and you can just export them as Onyx and
1:11:12
then Tiny Grad can run Onyx. So
1:11:15
the ports that I did of llama,
1:11:17
stable diffusion and now whisper
1:11:18
are more academic to teach me about
1:11:21
the models, but they are cleaner
1:11:23
than the PyTorch versions. You can read the code.
1:11:25
I think the code is easier to read. It's less lines.
1:11:28
There's just a few things about the way Tiny Grad writes
1:11:30
things. Here's a complaint I have about PyTorch. NN.relu
1:11:34
is a class,
1:11:36
right? So when you create an NN module, you'll
1:11:38
put your NN relu
1:11:41
as in a knit. And this makes
1:11:43
no sense. Relu is completely stateless.
1:11:46
Why should that be a class?
1:11:48
But that's more like a software engineering
1:11:51
thing. Or do you think it has a cost on performance?
1:11:53
Oh no, it doesn't have a cost on performance. But
1:11:56
yeah, no, I think that it's, that's
1:11:58
what I mean about Tiny Grad's front.
1:11:59
and being cleaner. I see.
1:12:03
What do you think about Mojo? I don't know if you've been paying attention
1:12:05
to the programming language that does some
1:12:08
interesting ideas that kind of intersect
1:12:10
TinyGrad.
1:12:11
I think that there is a spectrum. And
1:12:14
like on one side you have Mojo and on the other
1:12:16
side you have like GGML. GGML
1:12:19
is this like, we're gonna run llama fast on
1:12:21
Mac. Okay, we're gonna expand out
1:12:23
to a little bit, but we're gonna basically go like depth first,
1:12:26
right? Mojo is like, we're gonna go breath
1:12:28
first. We're gonna go so wide that we're gonna make all
1:12:30
of Python fast and TinyGrad's in the middle.
1:12:33
TinyGrad is, we are going
1:12:35
to make neural networks fast.
1:12:38
Yeah, but they try to really
1:12:41
get it to be fast, compiled
1:12:43
down to the specifics hardware
1:12:46
and make that compilation step
1:12:49
as flexible and resilient as possible.
1:12:51
Yeah, but they have Turing completeness.
1:12:53
And that limits you.
1:12:55
Turing. That's what you're saying, it's somewhere in
1:12:57
the middle. So you're actually going to be targeting some
1:12:59
accelerators, some number,
1:13:02
not one. My
1:13:05
goal is step one, build
1:13:07
an equally performance stack to PyTorch
1:13:09
on NVIDIA and AMD, but
1:13:12
with way less lines.
1:13:13
And then step two is, okay, how do
1:13:15
we make an accelerator? But you need step
1:13:17
one. You have to first build the framework before
1:13:20
you can build the accelerator. Can you explain
1:13:22
MLperf? What's your
1:13:24
approach in general to benchmarking TinyGrad performance?
1:13:27
So I'm much more
1:13:30
of a, like,
1:13:32
build it the right way and worry
1:13:34
about performance later.
1:13:36
There's a bunch of things where I haven't
1:13:38
even like,
1:13:39
really dove into performance. The only place
1:13:41
where TinyGrad is competitive performance wise right now
1:13:44
is on Qualcomm GPUs. So
1:13:46
TinyGrad's actually used an open pilot to run the model. So
1:13:49
the driving model is TinyGrad. When
1:13:51
did that happen? That transition?
1:13:53
About eight months ago now.
1:13:56
And it's 2x faster than Qualcomm's library. What's
1:13:59
the hardwood? where
1:14:01
that open pilot runs on the Kamaia.
1:14:04
It's a Snapdragon 845. Okay.
1:14:07
So this is using the GPU. So the GPU's an Adreno GPU.
1:14:10
There's like different things. There's a really good Microsoft
1:14:13
paper that talks about like mobile GPUs
1:14:15
and why they're different from desktop GPUs. One
1:14:18
of the big things is in a desktop
1:14:20
GPU, you can use buffers
1:14:22
on a mobile GPU image textures
1:14:24
a lot faster.
1:14:27
And a mobile GPU image textures, okay. And
1:14:30
so you want to be able to leverage
1:14:33
that.
1:14:34
I want to be able to leverage it in a way that it's completely
1:14:36
generic, right? So there's a lot of this. Xiaomi
1:14:38
has a pretty good open source library for
1:14:40
mobile GPUs called MACE, where
1:14:42
they can generate where they have these kernels,
1:14:45
but they're all hand coded, right? So
1:14:47
that's great if you're doing three by three comps. That's
1:14:49
great if you're doing dense map models. But the minute
1:14:51
you go off the beaten path a tiny bit, well, your
1:14:54
performance
1:14:54
is nothing. Since
1:14:56
you mentioned OpenPilot, I'd love to get an update
1:14:58
in the company number
1:15:01
one, Calm AI World. How are
1:15:03
things going there in the development of
1:15:07
semi-autonomous driving?
1:15:10
You know, almost no one talks
1:15:13
about FSD anymore, and even less people
1:15:15
talk about OpenPilot. We've solved
1:15:17
the problem. Like, we solved it years ago.
1:15:21
What's the problem exactly? Well,
1:15:23
how do you... What does solving it mean?
1:15:26
Solving means how do you build a model that
1:15:29
outputs a human policy for driving?
1:15:31
How do you build a model that, given a reasonable
1:15:34
set of sensors, outputs a human policy for driving?
1:15:37
So you have companies
1:15:39
like Waymo and Cruise, which are hand coding these things that
1:15:41
are like quasi human policies. Then
1:15:45
you have Tesla
1:15:47
and maybe even to more of an extent,
1:15:50
comma, asking, okay, how do we just learn the human
1:15:52
policy and data? The
1:15:55
big thing that we're doing now, and we just put it out on Twitter...
1:16:00
At the beginning of comma,
1:16:02
we published a paper
1:16:03
called Learning a Driving Simulator.
1:16:06
And the way this thing worked was it
1:16:09
was an auto encoder and
1:16:11
then an RNN in the middle. You
1:16:14
take an auto encoder, you compress
1:16:16
the picture,
1:16:17
you use an RNN, predict the next state, and
1:16:19
these things were. It was a laughably
1:16:22
bad simulator. This is
1:16:24
2015-hour machine learning technology. Today
1:16:26
we have VQ, VAE, and transformers.
1:16:29
We're building Drive GPT basically.
1:16:32
Drive GPT. It's
1:16:37
trained on what? Is it trained in a self-supervised
1:16:39
way? It's trained on all the driving data
1:16:41
to predict the next frame.
1:16:43
Really trying to learn
1:16:45
a human policy. What would a human do? Well,
1:16:48
actually our simulator is conditioned on the pose. It's
1:16:50
actually a simulator. You can put in a state action
1:16:52
pair and get out the next state.
1:16:54
And then once
1:16:56
you have a simulator, you can do RL
1:16:58
in the simulator and RL will get us that
1:17:01
human policy.
1:17:02
So transfers. Yeah.
1:17:05
RL with a reward function, not
1:17:07
asking is this close to the human policy, but asking
1:17:09
would a human disengage if you did this behavior?
1:17:12
Okay. Let me think about the distinction
1:17:15
there. What a human disengage. What
1:17:18
a human disengage. That
1:17:22
correlates, I guess, with human policy,
1:17:24
but it could be different. So
1:17:27
it doesn't just say what would a human
1:17:29
do. It says what
1:17:30
would a good human driver do and
1:17:32
such that the experience is comfortable,
1:17:36
but also not annoying in that the thing
1:17:38
is very cautious. So it's
1:17:41
finding a nice balance. That's interesting. It's
1:17:43
a nice... It's asking exactly the right question. What
1:17:46
will make our customers happy? Right.
1:17:49
A system that you never wanted to engage.
1:17:51
Because usually disengagement is almost
1:17:54
always a sign of I'm not
1:17:56
happy with what the system is doing. Usually.
1:17:59
There's some that are just I fell. like driving and those
1:18:01
are always fine too but they're just going to look like noise
1:18:03
in the data. But even
1:18:05
that felt like driving. Maybe
1:18:07
yeah. That's even that's a signal like why
1:18:09
do you feel like driving here you
1:18:12
need to recalibrate
1:18:14
your relationship with the car. Okay
1:18:17
so what that that's really interesting. How
1:18:20
close are we just solving self-driving?
1:18:25
It's hard to say.
1:18:26
We haven't completely closed the loop yet
1:18:29
so we don't have anything built that truly looks like
1:18:31
that architecture yet. We have prototypes
1:18:34
and there's bugs. So we are
1:18:36
a
1:18:36
couple bug fixes away. Might take
1:18:39
a year might take 10. What's the
1:18:41
nature of the bugs? Are these
1:18:44
these major philosophical bugs logical
1:18:46
bugs? What kind of what kind of bugs are we talking about?
1:18:48
They're just like they're just like stupid bugs and like
1:18:50
also we might just need more scale. We
1:18:52
just massively expanded our compute
1:18:54
cluster at Kama. We
1:18:57
now have about two people worth of compute 40
1:18:59
beta flops.
1:19:00
Well people people
1:19:03
are different. Yeah 20 beta
1:19:05
flops. That's a person. It's just a unit right. Horses
1:19:08
are different too but we still call it a horsepower. Yeah
1:19:11
but there's something different about mobility
1:19:13
than there is about
1:19:14
perception and action
1:19:17
in a very complicated world. But yes.
1:19:19
Well yeah of course not all flops are created equal. If you
1:19:21
have randomly initialized weights it's not gonna. Not
1:19:24
all flops are created equal. So flops
1:19:26
are doing way more useful things than others. Yeah.
1:19:31
Tell me about it. Okay so more
1:19:33
data scale means more scale in compute
1:19:35
or scale in scale of data?
1:19:37
Both. Diversity
1:19:41
of data? Diversity is very important in data.
1:19:43
Yeah I mean
1:19:45
we have so we have about I think
1:19:47
we have like 5 000 daily actives.
1:19:51
How would you evaluate how uh FSD
1:19:54
is doing? Pretty well.
1:19:56
How's that race going
1:19:58
between Kama AI and FSD?
1:19:59
Tesla has always wanted two years ahead of us. They've
1:20:02
always been wanted two years ahead of us. And they
1:20:04
probably always will be because they're not doing anything wrong. What
1:20:07
have you seen that's since the last time we talked that
1:20:09
are interesting architectural decisions, training decisions,
1:20:12
like the way the way they deploy stuff, the architectures
1:20:14
they're using in terms of the software,
1:20:16
how the teams are run, all that kind of stuff, data collection.
1:20:19
Anything interesting? I mean, I know they're moving
1:20:21
toward more of an end to end approach.
1:20:23
So creeping towards end to end
1:20:25
as much as possible across the
1:20:28
whole thing. The training, the data
1:20:30
collection, everything. They also have a very fancy simulator.
1:20:32
They're probably saying all the same things we are. They're
1:20:34
probably saying we just need to optimize, you know, what
1:20:37
is the reward? We get negative reward for this engagement.
1:20:39
Right? Like, everyone kind of knows this.
1:20:41
It's just a question who can actually build and deploy the system.
1:20:44
Yeah. I mean, this good, it's requires
1:20:46
good software engineering, I think. Yeah. And
1:20:49
the right kind of hardware. Yeah,
1:20:51
the hardware to run it. You
1:20:53
still don't believe in cloud in that regard?
1:20:57
I have a compute cluster
1:20:59
in my office. 800 amps. Tiny
1:21:02
grad. It's 40 kilowatts at idle,
1:21:04
our data center.
1:21:06
That's crazy. If 40 kilowatts is burning
1:21:08
just when the computers are idle.
1:21:09
Just when I... Oh, sorry. Sorry. Compute cluster. Compute
1:21:14
cluster. I got it. It's not a data center. Yeah. Now,
1:21:16
data centers are clouds. We don't have clouds.
1:21:19
Data centers have air conditioners. We have fans. That
1:21:22
makes it a compute cluster. I'm
1:21:25
guessing this is a kind of a legal distinction
1:21:27
as compared to me. Sure. Yeah. We have a compute
1:21:29
cluster.
1:21:31
You said that you don't think LLMs have consciousness,
1:21:33
or at least not more than a chicken.
1:21:36
Do you think they can reason? Is there something
1:21:38
interesting to you about the word reason, about
1:21:41
some of the capabilities that we think is kind of human,
1:21:43
to be able to
1:21:45
integrate
1:21:47
complicated information and through
1:21:50
a chain of thought arrive
1:21:54
at a conclusion that feels novel, a novel
1:21:57
integration of the... disparate
1:22:00
facts. Yeah,
1:22:03
I don't think that there's, I think that
1:22:05
they can reason better than a lot of people. Hey,
1:22:08
isn't that amazing to you though? Isn't
1:22:10
that like an incredible thing that a transformer
1:22:12
can achieve? I mean, I think
1:22:14
that calculators can add better than a lot
1:22:16
of people. But language feels
1:22:19
like reasoning through the process
1:22:21
of language, which
1:22:23
looks a lot like thought. Making
1:22:27
brilliant season chess, which feels
1:22:29
a lot like thought. Whatever new
1:22:31
thing that AI can do, everybody thinks is
1:22:33
brilliant. And then like 20 years go by and they're like,
1:22:35
well, you have a chess, that's like mechanical. Like adding,
1:22:37
that's like mechanical. So you think language is not
1:22:39
that special. It's like chess. It's like chess
1:22:42
and it's like- I don't know. Because it's very
1:22:44
human, we take it, listen,
1:22:47
there's something different between chess and
1:22:51
language. Chess is a game that a subset
1:22:53
of the population plays. Language is something
1:22:56
we
1:22:56
use nonstop for all
1:22:58
of our human interaction. And human interaction
1:23:01
is fundamental to society. So
1:23:03
it's like, holy shit. This
1:23:06
language thing is not so difficult to
1:23:08
like
1:23:09
create in the machine. The problem
1:23:12
is if you go back to 1960 and you
1:23:14
tell them that you have a machine that can play
1:23:17
amazing chess,
1:23:19
of course someone in 1960 will tell you that machine
1:23:21
is intelligent.
1:23:23
Someone in 2010 won't, what's changed,
1:23:25
right? Today, we think that these machines
1:23:27
that have language are intelligent. But
1:23:30
I think in 20 years, we're gonna be like, yeah, but can it
1:23:32
reproduce?
1:23:33
So reproduction, yeah,
1:23:36
we might redefine what it means to be,
1:23:39
what is it? A high performance living
1:23:41
organism on earth. Humans are always gonna
1:23:44
define a niche for themselves. Like, well,
1:23:46
you know, we're better than the machines because we can,
1:23:48
you know, and like they tried creative for a bit, but
1:23:50
no one believes that one anymore.
1:23:52
But niche, is
1:23:54
that delusional? There's some accuracy to that. Because
1:23:57
maybe like with chess, you start to realize like.
1:23:59
that
1:24:02
we have, it'll conceive notions of
1:24:05
what makes humans special.
1:24:07
Like the apex organism
1:24:09
on Earth. Yeah,
1:24:12
and I think maybe we're going to go through that same
1:24:14
thing with language.
1:24:16
And that same thing with creativity.
1:24:19
But
1:24:19
language carries these notions of truth
1:24:22
and so on. And so we might be like, wait,
1:24:24
maybe truth is not carried by language.
1:24:27
Maybe there's like a deeper thing. The niche is
1:24:29
getting smaller. Oh, boy. But
1:24:33
no, no, no, you don't understand humans are
1:24:35
created by God and machines are created
1:24:37
by humans, therefore. Right? Like that'll
1:24:39
be the last niche we have.
1:24:41
So what do you think about this,
1:24:43
the rapid development of LMS? If we could
1:24:45
just like stick on that. It's still incredibly
1:24:47
impressive, like with Chagibiti. Just even Chagibiti,
1:24:49
what are your thoughts about reinforcement
1:24:52
learning with human feedback on these large language
1:24:54
models?
1:24:55
I'd like to go back to when calculators
1:24:58
first came out
1:24:59
and or computers. And
1:25:02
like I wasn't around. Look, I'm 33 years old. And
1:25:05
to like
1:25:06
see how that affected
1:25:09
like
1:25:12
society.
1:25:13
Maybe you're right. I want to put on
1:25:16
the the big picture
1:25:18
hat here. Oh my God, a refrigerator? Wow.
1:25:21
Refrigerator, electricity, all that kind of stuff.
1:25:25
But no,
1:25:26
with the Internet,
1:25:28
large language models seeming human like
1:25:31
basically passing a Turing test. It
1:25:33
seems it might have really at scale
1:25:36
rapid transformative effects on society.
1:25:39
But you're saying like other technologies have as well.
1:25:43
So maybe calculators not the best
1:25:45
example that because that just seems
1:25:47
like a may. Well, no, maybe
1:25:49
calculator. The poor milk man, the day he
1:25:51
learned about refrigerators, he's like, I'm done. You
1:25:55
tell me you just keep the milk in your house. You
1:25:58
don't need to deliver it every day. I'm done.
1:26:00
Well, yeah, you have to actually look at the practical
1:26:02
impacts of certain technologies that they've
1:26:04
had. Yeah, probably electricity
1:26:07
is a big one and also how rapidly it spread.
1:26:10
Man, the internet is a big one. I do think it's different
1:26:12
this time though.
1:26:13
Yeah, it just feels like stuff- The initiative is getting
1:26:15
smaller. The initiative is
1:26:17
humans. Yes. That
1:26:20
makes humans special. Yes. It
1:26:22
feels like it's getting smaller rapidly though, doesn't
1:26:25
it? Or is that just a feeling we dramatize
1:26:27
everything? I think we dramatize everything. I
1:26:29
think that you asked the milk
1:26:32
man when he saw the refugee writers, and they're
1:26:34
going to have one of these in every home? Yeah,
1:26:38
yeah, yeah. Yeah,
1:26:41
but boy, is it impressive. So
1:26:44
much more impressive than seeing a
1:26:47
chess world champion AI system. I
1:26:49
disagree, actually.
1:26:51
I disagree. I
1:26:53
think things like Mu Zero and AlphaGo
1:26:55
are so much more impressive because
1:26:57
these things are playing beyond
1:27:00
the highest human level. The
1:27:03
language models are writing middle
1:27:06
school level essays, and people are like, wow,
1:27:08
it's a great essay. It's a great five-paragraph
1:27:10
essay about the causes of the Civil War. Okay,
1:27:13
forget the Civil War, just generating code, codex.
1:27:15
Oh. So you're saying it's
1:27:18
mediocre code. Terrible. But
1:27:20
I don't think it's terrible. I think it's just mediocre
1:27:23
code. Yeah.
1:27:25
Often
1:27:25
close to correct. Like
1:27:27
for mediocre purposes. That's the scariest kind
1:27:30
of code. I spend 5% of time typing and 95% of time debugging.
1:27:33
The last thing I want is close to correct code.
1:27:36
I want a machine that can help me with the debugging, not
1:27:38
with the typing. You know, it's like level
1:27:40
two driving, similar
1:27:42
kind of thing. Yeah, you still should
1:27:45
be a good programmer in order to modify.
1:27:48
I wouldn't even say debugging. It's just modifying
1:27:50
the code, reading it. Don't think it's like level
1:27:52
two driving.
1:27:54
I think driving is not tool complete and programming
1:27:56
is. Meaning you don't use like the best
1:27:58
possible tools to drive. You're
1:28:01
not like, like, like,
1:28:03
cars have basically the same interface for
1:28:05
the last 50 years. Computers have
1:28:07
a radically different interface. Okay. Can
1:28:09
you describe the concept of tool complete?
1:28:12
Yeah. So think about the difference between a car from 1980 and
1:28:15
a car from today. Yeah. No difference
1:28:17
really. It's got a bunch of pedals, it's got a steering wheel.
1:28:20
Great.
1:28:20
Maybe now it has a few ADAS features, but
1:28:23
it's pretty much the same car. All right. You
1:28:25
have no problem getting into a 1980 car and driving it. Take
1:28:28
a programmer today who spent their whole life doing JavaScript
1:28:31
and you put him in an Apple 2e prompt and
1:28:33
you tell him about the line numbers in basic.
1:28:36
But how do I insert
1:28:38
something between line 17 and 18? Oh,
1:28:41
wow.
1:28:42
But the,
1:28:45
so in tool you're putting in the programming
1:28:47
languages. So it's just the entirety stack
1:28:49
of the tooling. Exactly. So it's not just
1:28:51
like the IDs or something like this. It's everything.
1:28:54
Yes. It's IDEs, the languages, the runtimes.
1:28:56
It's everything. It's tool complete.
1:28:59
So like almost if, if,
1:29:01
if, if, if, if codex or
1:29:03
co-pilot are helping you, that
1:29:05
actually probably means that your framework or library is
1:29:08
bad and there's too much boilerplate in it.
1:29:12
Yeah. But don't you think
1:29:14
so much programming has boilerplate? Tinygrad
1:29:17
is now 2,700 lines
1:29:19
and it can run llama and stable
1:29:21
diffusion and all of this stuff
1:29:24
is in 2,700 lines. boilerplate
1:29:26
and abstraction
1:29:29
indirection and all these things are
1:29:31
just bad code. Well,
1:29:36
let's talk about good code and bad
1:29:38
code. There's a, I would
1:29:40
say, I don't know, for generic
1:29:42
scripts that I write just offhand, like
1:29:45
I, like 80% of it is written by GPT.
1:29:48
Just like quick, quick like offhand
1:29:50
stuff. So not like libraries, not like performing
1:29:53
code, not stuff for robotics and so
1:29:55
on. Just quick stuff because your basics,
1:29:57
so much of programming is doing some.
1:29:59
some, yeah, boilerplate, but
1:30:02
to do so efficiently and quickly,
1:30:06
because you can't really automate it fully with
1:30:08
like generic method, like a generic
1:30:11
kind of ID
1:30:13
type of recommendation or something like this,
1:30:15
you do need to have some of the complexity of
1:30:17
language models.
1:30:19
Yeah, I guess if I was really writing like
1:30:21
maybe today, if I wrote like a
1:30:23
lot of like data parsing stuff, I mean,
1:30:25
I don't play CTFs anymore, but if I still play CTFs,
1:30:27
a lot of the like, is just like you have to write like a parser for this
1:30:29
data format, like I wonder, or
1:30:32
like advent of code,
1:30:33
I wonder when the models are gonna
1:30:35
start to help with that kind of code,
1:30:37
and they may, they may, and the models
1:30:39
also may help you with speed, and the
1:30:41
models are very fast, but where
1:30:43
the models won't, my programming
1:30:46
speed is not at all limited by
1:30:48
my typing speed.
1:30:52
And in very few cases
1:30:54
it is, yes, if I'm writing some script to
1:30:56
just like parse some weird data format, sure,
1:30:59
my programming speed is limited by my typing speed. What about
1:31:01
looking stuff up? Because that's essentially
1:31:03
a more efficient look up, right? You know,
1:31:06
when I was at Twitter, I tried to use chat
1:31:09
GPT to like ask
1:31:11
some questions, like was the API for this? And
1:31:14
it would just hallucinate,
1:31:15
it would just give me completely made up API
1:31:18
functions that sounded real. Well,
1:31:20
do you think that's just a temporary kind of stage?
1:31:23
No. You don't think it'll
1:31:25
get better and better and better and this kind of stuff, because like
1:31:27
it only hallucinates stuff in the edge
1:31:29
cases. Yes.
1:31:30
If you're writing generic code, it's actually pretty good. Yes,
1:31:32
if you are writing an absolute basic like
1:31:34
React app with a button, it's not gonna hallucinate,
1:31:36
sure. No, there's kind of
1:31:38
ways to fix the hallucination problem. I think Facebook
1:31:41
has an interesting paper, it's called Atlas, and
1:31:43
it's actually weird the way that we do language
1:31:46
models right now where all of the
1:31:49
information is in the weights.
1:31:51
And human brains don't really like this. It's
1:31:53
like a hippocampus and a memory system. So
1:31:55
why don't LLMs have a memory system? And there's
1:31:57
people working on them. I think future LLMs are gonna.
1:31:59
be smaller, but
1:32:02
are going to run looping
1:32:04
on themselves and are going to have retrieval systems.
1:32:07
And the thing about using a retrieval system is you can
1:32:09
cite sources,
1:32:10
explicitly. Which
1:32:14
is really helpful to integrate
1:32:16
the human into the loop of the
1:32:18
thing, because you can go check the sources and you can investigate.
1:32:21
So whenever the thing is hallucinating, you can
1:32:23
have the human supervision. That's pushing
1:32:26
it towards level two kind of drive. That's gonna kill Google.
1:32:29
Wait, which part? When someone makes an LLM
1:32:31
that's capable of citing its sources, it will kill Google.
1:32:34
LLM that's citing its sources because that's basically
1:32:37
a search engine.
1:32:38
That's what people want in a search engine. But
1:32:40
also Google might be the people that build it. Maybe.
1:32:43
And put ads on it. I'd count them out. Why
1:32:46
is that? Why do you think? Who wins
1:32:48
this race? We got,
1:32:51
who are the competitors? We got
1:32:54
TinyCorp. I don't know if that's, yeah,
1:32:57
I mean, you're a legitimate competitor in that.
1:32:59
I'm not trying to compete on that. You're not.
1:33:02
No, not as a skit. It's gonna accidentally stumble into that
1:33:04
competition. Maybe. to
1:33:07
replace Google search?
1:33:08
When I started Comma, I said,
1:33:11
over and over again, I'm going to win self-driving cars.
1:33:13
I still believe that.
1:33:15
I have never said I'm going to win search
1:33:17
with the TinyCorp and
1:33:19
I'm never going to say that because I won't. The night
1:33:21
is still young. You don't know how
1:33:23
hard is it to win search in
1:33:25
this new route. It
1:33:28
feels, I mean, one of the things that ChatGPT kind of shows
1:33:30
that there could be a few interesting tricks that
1:33:32
really have, that create a really compelling product. Some
1:33:35
startup's gonna figure it out. I think
1:33:37
if you ask me, like Google's still the number one
1:33:39
webpage, I think by the end of the decade, Google won't be
1:33:41
the number one webpage anymore.
1:33:43
So you don't think Google, because of
1:33:45
the, how big the corporation is?
1:33:47
Look, I would put a lot more money on Mark Zuckerberg.
1:33:50
Why is that? Because
1:33:53
Mark Zuckerberg's alive. Like
1:33:57
this is old Paul Graham essay. Startups are
1:33:59
either alive or dead. Google's dead.
1:34:02
Facebook's alive. Versus Facebook is alive,
1:34:04
Meta is alive. Meta. Meta. You
1:34:06
see what I mean? Like that's just, like Mark
1:34:08
Zuckerberg, this is Mark Zuckerberg reading that Paul Graham
1:34:10
asking and being like, I'm gonna show everyone how alive
1:34:12
we are. I'm gonna change the name.
1:34:14
So you don't think there's this gutsy
1:34:18
pivoting engine that,
1:34:22
like Google doesn't have that, the kind of engine
1:34:24
that a startup has like constantly being
1:34:27
alive, I guess. When I listened to your Sam
1:34:29
Altman
1:34:30
podcast, he talked about the button. Everyone
1:34:32
who talks about AI talks about the button, the button to turn it off,
1:34:34
right? Do we have a button to turn off Google?
1:34:37
Is anybody
1:34:39
in the world capable of shutting Google down? What
1:34:43
does that mean exactly? The company or the search
1:34:45
engine? Could we shut the search engine down? Could we shut the company
1:34:47
down? Either. Can
1:34:50
you elaborate on the value of that question? Does
1:34:52
Sundar Pichai have the authority to turn
1:34:54
off google.com tomorrow? Who
1:34:57
has the authority? That's a good question. Does
1:35:00
anyone? Does anyone? Yeah, I'm sure.
1:35:03
Are you sure?
1:35:04
No, they have the technical power, but do they
1:35:06
have the authority? Let's say Sundar Pichai made
1:35:09
this his sole mission. He came into Google
1:35:11
tomorrow and said, I'm gonna shut google.com down.
1:35:14
I don't think he'd keep his position too long. And
1:35:18
what is the mechanism by which he wouldn't keep his position?
1:35:21
Well, boards and shares
1:35:23
and corporate undermining and oh my
1:35:25
God, our revenue is zero now. Okay,
1:35:29
so what's the case you're making here? So the
1:35:31
capitalist machine prevents you from having
1:35:34
the button. Yeah,
1:35:35
and it will have, I mean, this is true for the AIs too. There's
1:35:38
no turning the AIs off.
1:35:40
There's no button. You can't press it. Now,
1:35:42
does Mark Zuckerberg have that button for Facebook.com?
1:35:46
Yes, probably more. I think he does. I
1:35:49
think he does, and this is exactly what I mean
1:35:51
and why I bet on him so much more than
1:35:53
I bet on Google. I guess you could say Elon
1:35:55
has similar stuff. Oh, Elon has the button.
1:35:59
Yeah.
1:36:00
Does he want can you on fire the missiles? Can
1:36:02
he fire the missiles? I
1:36:04
think some questions are better I'm
1:36:07
asked I mean,
1:36:09
you know a rocket an ICB. Yeah, well your rocket
1:36:11
that can land anywhere. Is that an ICB? M? Well,
1:36:14
yeah, you know don't ask too many questions my
1:36:17
god
1:36:19
But the
1:36:21
the positive side of the button is that you can
1:36:23
innovate aggressively is what you're saying
1:36:25
which is what's required with Turning
1:36:28
LLM into a search engine. I would bet on a startup.
1:36:31
I bet is it so easy, right? I bet on something that looks
1:36:33
like mid-journey but for search
1:36:37
Just is able to set sources loop
1:36:39
on itself, I mean just feels like one model can take off
1:36:41
Yeah, right and that nice wrapper and some
1:36:43
of it scale. I mean, it's hard to Like
1:36:46
create a product that just works really nicely Stably
1:36:49
the other thing that's gonna be cool is there is
1:36:51
some aspect of a winner take all effect, right?
1:36:54
Like once um
1:36:56
Someone starts deploying a product that gets a lot
1:36:58
of usage and you see this with open AI They
1:37:00
are going to get the data set
1:37:02
to train future versions of the model
1:37:04
Yeah, they are going to be able to right, you
1:37:07
know I was asked a Google image search when I worked there like
1:37:09
almost 15 years ago now How does Google know which
1:37:11
image is an apple
1:37:12
and I said the metadata and they're like, yeah that
1:37:14
works about half the time How does Google know
1:37:16
you'll see the role apples on the front page when you search Apple?
1:37:19
Mm-hmm. And I don't know.
1:37:21
I didn't come up with the answer
1:37:22
The guys like what's what people click on when they search Apple? Yeah,
1:37:26
yeah that data is really really powerful. It's the
1:37:28
human supervision What do you think
1:37:30
are the chances? What do you think in general
1:37:33
that llama was open-sourced? I just
1:37:36
did a conversation with With
1:37:38
Mark Zuckerberg and he's all
1:37:41
in on open source
1:37:43
Who would have thought that Mark Zuckerberg
1:37:45
would be the good guy? I Mean
1:37:47
it Would have thought anything
1:37:50
in this world It's hard to know
1:37:53
but open source to you ultimately
1:37:57
Is a good thing here undoubtedly
1:38:01
You know, what's ironic
1:38:03
about all these AI safety people is
1:38:05
they are going to build the exact thing they fear.
1:38:09
These we need to have one model that we
1:38:11
control and align. This is
1:38:13
the only way you end up paper clipped. There's
1:38:16
no way you end up paper clipped if everybody
1:38:18
has an AI. So open sourcing is
1:38:20
the way to fight the paper clip, Maximizer? Absolutely.
1:38:23
It's the only way. You think you're going to control
1:38:26
it? You're not going to control it. So the
1:38:28
criticism you have for the AI safety folks
1:38:31
is that there is a belief
1:38:33
and a desire for control. And
1:38:36
that belief and desire for centralized
1:38:38
control of dangerous AI systems
1:38:41
is not good. Sam Altman won't tell you
1:38:43
that GPT-4 has 220 billion
1:38:46
parameters and is a 16-way mixture model
1:38:48
with eight sets of weights.
1:38:50
Who did you have to murder to
1:38:52
get that information? All right. I
1:38:54
mean, look. But yes. Everyone
1:38:57
at OpenAI knows what I just said was true. Now
1:39:01
ask the question, really.
1:39:03
It upsets me when I like GPT-2.
1:39:06
When OpenAI came out with GPT-2 and raised a whole
1:39:08
fake AI safety thing about that, I mean, now the
1:39:10
model is laughable.
1:39:12
They used AI safety
1:39:14
to hype up their company and it's disgusting.
1:39:18
Or the flip side of that is
1:39:21
they used a relatively weak model
1:39:23
in retrospect to explore how
1:39:25
do we do AI safety correctly?
1:39:27
How do we release things? How do we go through the process? I
1:39:30
don't know if... Sure. Sure.
1:39:33
All right. All right. That's
1:39:35
the charitable interpretation. I don't know how much hype there is in AI safety,
1:39:37
honestly. Oh, there's so much. At least on Twitter.
1:39:40
I don't know. Maybe Twitter's not real life. Twitter's
1:39:42
not real life. Come on. In
1:39:44
terms of hype. I mean, I don't... I
1:39:47
think OpenAI has been finding an
1:39:49
interesting balance between transparency
1:39:50
and putting value on
1:39:53
AI safety. You
1:39:55
think just go all out open
1:39:57
source. So do a llama. Absolutely.
1:40:00
So do like open source, this
1:40:02
is a tough question, which is open
1:40:05
source, both the base, the
1:40:07
foundation model and the fine tune
1:40:09
one. So like the
1:40:11
model that can be ultra racist and dangerous
1:40:14
and like tell you how to build
1:40:16
a nuclear weapon. Oh my God, have you met humans,
1:40:19
right? Like half of these AI- I haven't
1:40:21
met most humans. This makes,
1:40:23
this allows you to meet every human.
1:40:26
Yeah, I know, but half of these AI alignment
1:40:28
problems are just human alignment problems. And
1:40:30
that's what's also so scary about the language they
1:40:32
use. It's like, it's not the machines you want to align,
1:40:34
it's me.
1:40:37
But here's the thing, it
1:40:39
makes it very accessible to ask
1:40:43
very questions where
1:40:46
the answers have dangerous consequences if
1:40:48
you were to act on them. I
1:40:51
mean, yeah, welcome to the world.
1:40:54
Well, no, for me, there's a lot of friction. If
1:40:56
I want to find out how to,
1:40:59
I don't know, blow
1:41:01
up something. No, there's not a lot of friction that's
1:41:03
so easy. No, like what do I search
1:41:06
that is Bing? Or do I search anything that
1:41:08
I use? No, there's like lots of stuff. No,
1:41:10
it feels like I have to keep clicking a lot of this. First off, first off,
1:41:13
first off, anyone who's stupid enough to search for how
1:41:15
to blow up a building in my neighborhood
1:41:18
is not smart enough to build a bomb, right?
1:41:20
Are you sure about that? Yes. I
1:41:24
feel like a language model makes it
1:41:26
more accessible for that
1:41:29
person who's not smart enough to do. They're
1:41:31
not gonna build a bomb, trust me. The
1:41:34
people who are incapable of figuring
1:41:36
out how to ask that question a bit more academically
1:41:39
and get a real answer from it are not capable
1:41:41
of procuring the materials, which are somewhat controlled
1:41:43
to build a bomb. No, I think
1:41:45
it all makes it more accessible to people with money
1:41:48
without the technical know-how,
1:41:50
right? Like, do you really
1:41:52
need to know how to build a bomb to build a bomb? You
1:41:54
can hire people, you can find like- Or you can hire
1:41:57
people to build a- You know what? I was asking this question
1:41:59
on my stream. Like, can Jeff-
1:41:59
Bezos hire a hit man probably not but
1:42:03
a language model can
1:42:05
probably help you out yeah you'll
1:42:07
still go to jail right like it's not like the language
1:42:10
model is God like the language model it's like it's
1:42:12
you literally just hired someone on Fiverr
1:42:15
but you use it but okay GPT
1:42:17
for in terms of finding hit man is like asking
1:42:19
Fiverr how to find a I understand but
1:42:21
don't you think you how you know okay how but
1:42:23
don't you think GPT 5 will be better just
1:42:26
don't you think that information is out there on the internet I
1:42:29
mean yeah
1:42:29
and I think that if someone is actually
1:42:31
serious enough to hire a hit man or build a bomb
1:42:34
they'd also be serious enough to find the information
1:42:36
I don't think so I think it makes it more accessible
1:42:38
if you have if you have enough money to buy
1:42:40
hit man I think it decreases
1:42:43
the friction of how hard is it to find
1:42:45
that kind of hit man I honestly think this
1:42:48
there's a jump in
1:42:51
ease and scale of how
1:42:53
much harm you can do and I don't mean harm with language
1:42:56
I mean harm with actual violence what you're basically
1:42:58
saying is like okay what's gonna happen is these people
1:43:00
who are not intelligent are going to use
1:43:02
machines to augment their
1:43:04
intelligence and now intelligent people
1:43:07
and machines intelligence is scary intelligent
1:43:10
agents are scary when I'm in the
1:43:12
woods the scariest animal to meet is human
1:43:14
right
1:43:15
no no no there's look there's like nice California
1:43:18
humans like I see you're wearing like you
1:43:20
know street clothes and Nikes are fine
1:43:23
you look like you've been a human who's been in the woods for a while yeah
1:43:25
I'm more scared of you than a bear that's what they say about
1:43:27
the Amazon you go to the Amazon it's
1:43:30
the human tribes so
1:43:32
intelligence is scary
1:43:34
right so to just like ask this question
1:43:36
in generic way you're like what if we took
1:43:38
everybody who you know maybe has ill
1:43:41
intention but is not so intelligent and gave them intelligence
1:43:45
right so we
1:43:47
should have intelligence control of course we
1:43:49
should only give intelligence to good people and that
1:43:51
is the absolutely horrifying idea should you
1:43:54
the best defense is actually the best defense
1:43:56
is to give more intelligence to the
1:43:58
good guys and intelligent
1:43:59
Give intelligence to everybody. Give intelligence to everybody.
1:44:02
You know what, it's not even like guns, right? Like people say this about guns. You
1:44:04
know, what's the best defense against a bad guy with a gun, a good guy with a
1:44:06
gun? I'm like, I kinda subscribe to that, but I really
1:44:08
subscribe to that with intelligence.
1:44:10
Yeah, in a fundamental way, I agree
1:44:13
with you, but there just feels like so
1:44:15
much uncertainty and so much can happen rapidly that
1:44:17
you can lose a lot of control and you can do a lot of damage.
1:44:20
Oh no, we can lose control? Yes, thank
1:44:23
God. Yeah.
1:44:24
I hope they lose control.
1:44:28
I'd want them to lose control more than anything else. I
1:44:31
think when you lose control, you can do a lot of damage,
1:44:33
but you can do more damage when you centralize
1:44:36
and hold onto control is the point. Centralized
1:44:39
and held control is tyranny, right?
1:44:41
I will always, I don't like anarchy either, but
1:44:43
I've always taken anarchy over tyranny. Anarchy, you have
1:44:45
a chance. This
1:44:47
human civilization we've got going on is
1:44:50
quite interesting. I mean, I agree with you. So
1:44:52
do you open source is
1:44:55
the way forward here? So you admire what Facebook
1:44:57
is doing here or what Meta is doing with the release
1:44:59
of them. A lot. I lost $80,000 last year investing in
1:45:01
Meta and
1:45:04
when they released Llama, I'm like, yeah, whatever man, that
1:45:06
was worth it. That's
1:45:07
worth it. Do you think Google
1:45:09
and OpenAI with Microsoft
1:45:12
will match what Meta is doing
1:45:14
or not? So if
1:45:16
I were a researcher, why would you wanna
1:45:18
work at OpenAI? Like, you know, you're just,
1:45:21
you're on the bad team. Like, I mean
1:45:23
it, like you're on the bad team who can't even say that GPT-4
1:45:25
has 220 billion parameters. So close
1:45:27
source to use the bad team.
1:45:29
Not only close source, I'm not saying you need
1:45:31
to make your model weights open. I'm
1:45:33
not saying that. I totally understand we're keeping
1:45:36
our model weights closed because that's our product, right?
1:45:38
That's fine.
1:45:39
I'm saying like,
1:45:41
because of AI safety reasons, we can't
1:45:43
tell you the number of
1:45:44
billions of parameters in the model.
1:45:46
That's just the bad guys. Just
1:45:49
because you're mocking AI safety doesn't mean
1:45:51
it's not real. Oh, of course. Is it
1:45:53
possible that these things can really do a lot
1:45:55
of damage that we don't know? Oh my God,
1:45:57
yes. Intelligence is so dangerous.
1:45:59
human intelligence or machine intelligence.
1:46:02
Intelligence is dangerous. Machine
1:46:04
intelligence is so much easier to deploy at scale,
1:46:07
rapidly.
1:46:08
Okay, if you have human-like
1:46:10
bots on Twitter,
1:46:13
and you have a thousand of them, create
1:46:16
a whole narrative, like you can
1:46:19
manipulate millions of people.
1:46:21
But you mean like the intelligence agencies in America
1:46:23
are doing right now? Yeah, but they're not doing it
1:46:25
that well. It feels like you can do
1:46:27
a lot. They're doing it pretty well. Well,
1:46:31
I think they're doing a pretty good job. I suspect
1:46:34
they're not nearly as good as a bunch of GPT-fueled
1:46:37
bots could be. Well, I mean, of course they're looking
1:46:39
into the latest technologies for control of people,
1:46:41
of course.
1:46:42
But I think there's a George Hotz type character
1:46:44
that can do a better job than the entirety of them.
1:46:47
You don't think so? No way. No, and I'll tell you
1:46:49
why the George Hotz character can't. And I thought about this a lot with
1:46:51
hacking,
1:46:52
right? Like I can find exploits in web browsers. I probably still can.
1:46:54
I mean, I was better out on I was 24, but
1:46:56
the thing that I lack is the ability to
1:46:59
slowly and steadily deploy them over five years. And
1:47:01
this is what intelligence agencies are very good at.
1:47:04
Intelligence agencies don't have the most sophisticated
1:47:06
technology. They just
1:47:08
have- Endurance? Endurance.
1:47:12
Yeah, the financial backing and
1:47:15
the infrastructure for the endurance.
1:47:17
So the more we can decentralize
1:47:19
power, like you could make an argument by
1:47:22
the way that nobody should have these things. And
1:47:24
I would defend that argument. I would, like you're
1:47:26
saying that, look, LLMs and AI
1:47:28
and machine intelligence can cause a lot of harm,
1:47:31
so nobody should have it.
1:47:32
And I will respect someone philosophically
1:47:34
with that position. Just like I will respect someone philosophically
1:47:37
with a position that nobody should have guns,
1:47:39
right? But I will not respect philosophically
1:47:42
with only the trusted
1:47:44
authorities should have access to this.
1:47:47
Who are the trusted authorities? You know what? I'm
1:47:50
not worried about alignment between AI company
1:47:54
and their machines. I'm worried about alignment
1:47:56
between me and AI company.
1:47:58
What do you think?
1:47:59
as Eliot Kowski would say to you.
1:48:03
Because he is really against open source.
1:48:05
I know. And
1:48:09
I thought about this. I thought about this. And
1:48:13
I think this comes down to a
1:48:16
repeated misunderstanding of political
1:48:18
power by the rationalists.
1:48:21
Interesting. I
1:48:24
think that Eliot Kowski
1:48:26
is scared of these things. And I
1:48:28
am scared of these things too. Everyone
1:48:30
should be scared of these things. These things are scary.
1:48:33
But now you ask
1:48:35
about the two possible futures.
1:48:37
One where a small
1:48:39
trusted centralized group of people
1:48:41
has them. And the other where everyone has them.
1:48:44
And I am much less scared of the second
1:48:46
future than the first.
1:48:49
Well, there's a small trusted group of people that have
1:48:51
control of our nuclear weapons.
1:48:54
There's a difference.
1:48:55
Again, a nuclear weapon cannot be deployed
1:48:58
tactically. And a nuclear weapon is not a defense against
1:49:00
a nuclear weapon.
1:49:03
Except maybe in some philosophical mind game kind
1:49:05
of way.
1:49:06
But AI is different
1:49:09
how exactly? OK. Let's
1:49:11
say the
1:49:12
intelligence agency deploys a million bots
1:49:15
on Twitter or a thousand bots on Twitter to try to convince
1:49:17
me of a point.
1:49:19
Imagine I had a powerful AI running
1:49:21
on my computer saying, OK,
1:49:23
nice PSYOP. Nice PSYOP. Nice PSYOP.
1:49:26
OK. Here's a PSYOP. I filtered
1:49:28
it out for you.
1:49:29
Yeah. I mean, so you have fundamentally
1:49:32
hope for that, for
1:49:34
the defense of PSYOP. I'm not
1:49:36
even like, I don't even mean these things in truly horrible
1:49:38
ways. I mean these things in straight up ad blocker.
1:49:41
Right? Yeah. Straight up ad blocker. I don't want ads.
1:49:44
Yeah. But they are always finding, imagine
1:49:46
I had an AI that could just block
1:49:48
all the ads for me. So
1:49:50
you believe in the power
1:49:52
of the people to always create an ad blocker.
1:49:55
Yeah. I mean, I kind of share that belief.
1:49:58
I have, that's one of the. the deepest
1:50:01
optimism I have is just like, there's a lot
1:50:03
of good guys. So to
1:50:05
give, you shouldn't hand
1:50:07
pick them, just throw out powerful
1:50:10
technology out there and the good guys
1:50:12
will outnumber and out power
1:50:14
the bad guys. Yeah, I'm not even gonna say there's a
1:50:16
lot of good guys. I'm saying that good outnumber's bad,
1:50:19
right? Good outnumber's bad. In skill and performance.
1:50:22
Yeah, definitely in skill and performance, probably just in
1:50:24
number too. Probably just in general. I mean,
1:50:26
if you believe philosophically in democracy, you obviously
1:50:28
believe that.
1:50:30
That good outnumber's bad. And
1:50:33
like the only,
1:50:35
if you give it to a small number of people,
1:50:38
there's a chance you gave it to good people, but there's also a chance
1:50:41
you gave it to bad people. If you give it to everybody,
1:50:44
well if good outnumber's bad, then you definitely gave it
1:50:46
to more good people than bad.
1:50:47
That's
1:50:51
really interesting. So that's on the safety grounds, but then
1:50:53
also of course there's other motivations
1:50:55
like you don't wanna give away your secret sauce.
1:50:57
Well that's what I mean. I mean, I look, I respect capitalism.
1:51:00
I don't think that, I think that it would be polite
1:51:03
for you to make model architectures open source
1:51:05
and fundamental breakthroughs open source.
1:51:07
I don't think you have to make weights open source. You know what's interesting
1:51:10
is that
1:51:11
like there's so many possible trajectories
1:51:13
in human history where
1:51:16
you could have the next Google
1:51:18
be open source. So for example, I don't
1:51:20
know if that connection is accurate,
1:51:23
but you know, Wikipedia made a lot of interesting decisions,
1:51:25
not to put ads.
1:51:27
Wikipedia is basically open source.
1:51:29
You could think of it that way. And
1:51:31
like that's one of the main websites on the
1:51:33
internet. And like it didn't have to be that way.
1:51:36
It could have been like Google could have created Wikipedia,
1:51:38
put ads on it. You could probably run amazing
1:51:40
ads now on Wikipedia.
1:51:42
You wouldn't have to keep asking for money, but
1:51:45
it's interesting, right? So llama, open
1:51:47
source llama, derivatives of open
1:51:50
source llama might win the internet. I
1:51:53
sure hope so. I hope to see another
1:51:55
era. The kids today
1:51:58
don't know how good the internet used to be. And
1:52:00
I don't think this is just, oh, come on, like everyone's nostalgic
1:52:03
for their past, but I actually think
1:52:05
the internet, before small
1:52:07
groups of weaponized corporate and government
1:52:09
interests took it over, was a beautiful place.
1:52:15
You know, those small
1:52:17
number of companies have created some sexy
1:52:20
products, but you're saying
1:52:22
overall,
1:52:23
in the long arc of history, the
1:52:25
centralization of power they have like
1:52:28
suffocated the human spirit at scale. Here's
1:52:30
a question to ask about those beautiful, sexy
1:52:32
products. Imagine 2000 Google to 2010
1:52:35
Google, right? A lot
1:52:37
changed. We got maps, we got Gmail.
1:52:40
We lost a lot of products too, I think. Yeah,
1:52:42
I mean, some were probably, we've got Chrome, right?
1:52:44
And now let's go from 2010, we got Android. Now
1:52:47
let's go from 2010 to 2020. Well,
1:52:50
what does Google have? Well, search engine, maps,
1:52:53
mail, Android and Chrome. Oh,
1:52:55
I see. The internet
1:52:58
was this, you know, I was Times
1:53:00
person of the
1:53:00
year in 2006. I
1:53:04
love this. It's you, was Times person
1:53:06
of the year in 2006, right? Like that's,
1:53:09
you know, so quickly did people forget.
1:53:12
And I think some of it's social
1:53:14
media. I think some of it, I hope,
1:53:17
look, I hope that, I don't,
1:53:19
it's possible that some very sinister things happen.
1:53:22
I don't know. I think it might just be like the effects
1:53:24
of social media.
1:53:26
But something happened
1:53:28
in the last 20 years.
1:53:30
Oh, okay, so you're just being an old
1:53:32
man who's worried about the, I think there's always, it
1:53:34
goes, it's the cycle thing, there's ups and downs. And
1:53:36
I think people rediscover the power of distributed,
1:53:39
of decentralized. Yeah. I
1:53:41
mean, that's kind of like what the whole cryptocurrency is trying
1:53:44
to think that,
1:53:45
I think crypto is just carrying
1:53:47
the flame of that spirit of like, stuff should
1:53:49
be decentralized. It's just such a shame that they
1:53:51
all got rich,
1:53:53
you know? Yeah. If you could call the money
1:53:55
out of crypto, it would have been a beautiful place. Yeah.
1:53:58
But no, I mean, these people, you know. They sucked
1:54:01
all the value out of it and took it. Yeah,
1:54:04
money kind of corrupts the mind somehow.
1:54:06
It becomes a drug. You corrupted all
1:54:08
of crypto. You had coins worth billions of dollars
1:54:11
that had zero use.
1:54:15
You still have hope for crypto? Sure. I
1:54:17
have hope for the ideas. I really do. Yeah,
1:54:21
I mean, you know,
1:54:24
I want the US dollar to collapse. I
1:54:27
do. George Hawts.
1:54:31
Well, let me sort of on the AISAT,
1:54:33
do you think there's some interesting questions there, though,
1:54:37
to solve for the open source community in this case? So
1:54:39
like alignment, for example, or the
1:54:42
control problem. Like if you really
1:54:44
have super powerful, you said it's scary.
1:54:47
What do we do with it? So not control,
1:54:49
not centralized control, but like
1:54:51
if you were then you're gonna see
1:54:53
some guy or
1:54:55
gal release a super
1:54:57
powerful language model, open source. And
1:54:59
here you are, George Hawts, thinking, holy
1:55:02
shit. Okay, what ideas do I have to
1:55:05
combat this thing? So
1:55:08
what ideas would you have? I am
1:55:11
so much not worried about the
1:55:13
machine independently doing harm.
1:55:16
That's what some of these AI safety people seem
1:55:18
to think. They somehow seem to think that the machine,
1:55:20
like independently is gonna rebel against its
1:55:22
creator. So you don't think you'll find autonomy?
1:55:25
No, this is sci-fi B
1:55:27
movie garbage. Okay, what if
1:55:29
the thing writes code, basically writes
1:55:31
viruses? If
1:55:34
the thing writes viruses, it's
1:55:36
because the human told
1:55:39
it to write viruses. Yeah, but there's some things you can't
1:55:41
like put back in the box, that's kind of the whole
1:55:43
point, is it kind of spreads. Give it
1:55:45
access to the internet, it spreads, installs
1:55:47
itself,
1:55:48
modifies your shit. B, B, B,
1:55:50
B plot sci-fi, not real.
1:55:53
I'm trying to work, I'm trying to get better at my plot
1:55:55
writing. The thing that worries me,
1:55:57
I mean, we have a real danger to discuss,
1:55:59
and that is.
1:55:59
is bad humans using
1:56:02
the thing to do whatever bad, unaligned
1:56:04
AI thing you want. But this goes
1:56:06
to your previous concern
1:56:08
that who gets to define who's a good human, who's
1:56:10
a bad human? Nobody does, we give it to everybody.
1:56:13
And if you do anything besides give it to everybody,
1:56:15
trust me, the bad humans will get it.
1:56:18
Because that's who gets power. It's always the bad humans
1:56:20
who get power. Okay, power.
1:56:23
And
1:56:24
power turns even slightly good
1:56:26
humans to bad. Sure. That's the intuition
1:56:28
you have. I don't know.
1:56:31
I don't think everyone, I don't think everyone.
1:56:33
I just think that like,
1:56:35
here's the saying that I put in one of my
1:56:37
blog posts. When I was in the hacking
1:56:39
world,
1:56:40
I found 95% of people to be good
1:56:42
and 5% of people to be bad.
1:56:44
Like just who I personally judged as good people and bad people.
1:56:46
Like they believed about like good things for the world. They
1:56:49
wanted like flourishing and they wanted, you
1:56:51
know, growth and they wanted things like consider good,
1:56:53
right?
1:56:55
I came into the business world with Kama and I found the
1:56:57
exact opposite. I found 5% of
1:56:59
people good and 95% of people bad. I
1:57:01
found a world that promotes psychopathy. I
1:57:04
wonder what that means. I wonder if that
1:57:06
care, like, I
1:57:08
wonder if that's anecdotal or if it, if
1:57:12
there's true to that, there's something about capitalism
1:57:16
at the core that promotes the
1:57:18
people that run capitalism that promotes psychopathy.
1:57:21
That saying may of course be my own biases, right?
1:57:23
That may be my own biases that these people are a lot more
1:57:26
aligned with me than these other people.
1:57:28
Right? Yeah. So, you know, I
1:57:30
can certainly recognize that,
1:57:33
but you know, in general, I mean, this is like the
1:57:35
common sense maxim, which is the people
1:57:38
who end up getting power are never the ones you want with
1:57:40
it.
1:57:41
But do you have a concern of super
1:57:43
intelligent AGI,
1:57:46
open sourced, and then
1:57:48
what do you do with that? I'm not saying control
1:57:50
it, it's open source. What do we do with this
1:57:52
human species? That's not up to me. I
1:57:54
mean, you know, like I'm not a central planner.
1:57:56
No, not central planner, but you'll probably tweet as
1:57:59
a few days. left to live for the human species. I have
1:58:01
my ideas of what to do with it, and everyone else has their
1:58:04
ideas of what to do and make the best ideas win. But
1:58:06
at this point, do you brainstorm?
1:58:08
Because it's not regulation,
1:58:11
it could be decentralized regulation, where people agree
1:58:13
that this is just like, we create
1:58:15
tools that make it more difficult for
1:58:17
you
1:58:19
to maybe
1:58:22
make it more difficult for code to spread,
1:58:25
antivirus software, this kind of thing. But this is- You're
1:58:27
saying that you should build AI firewalls? That sounds good. You shouldn't
1:58:29
be running an AI firewall. Yeah, right, exactly. You should be running
1:58:32
an AI firewall to your mind.
1:58:34
You're constantly under- That's such an interesting idea.
1:58:37
Info wars, man. I don't
1:58:39
know if you're being sarcastic or not, but
1:58:41
I think there's power to that. It's like,
1:58:44
how do I protect my mind
1:58:48
from influence of human-like
1:58:50
or superhuman intelligent bots? I
1:58:53
would pay so much money for that product. I would
1:58:55
pay so much money for that product.
1:58:57
You know how much money I'd pay just for a spam
1:58:59
filter that works? Well,
1:59:01
on Twitter sometimes I would
1:59:03
like to have a protection
1:59:06
mechanism for my mind from the outrage mobs
1:59:10
because
1:59:12
they feel like bot-like behavior. It's
1:59:14
a large number of people that will just grab
1:59:17
a viral narrative
1:59:18
and attack anyone else that believes otherwise. And
1:59:20
it's like-
1:59:21
Whenever someone's telling me some story from the news,
1:59:23
I'm always like, I don't want to hear it, CIA op, bro, it's a
1:59:25
CIA op, bro. It doesn't matter if that's true or
1:59:27
not. It's just trying to influence your mind. You're
1:59:29
repeating an ad to me.
1:59:31
The viral mobs, is it like, yeah,
1:59:34
they're- To me, a defense against those mobs
1:59:37
is just getting multiple perspectives
1:59:40
always
1:59:41
from sources that make you feel kind of
1:59:45
like you're getting smarter. And
1:59:47
just actually just basically feels good. Like
1:59:50
a good documentary just feels
1:59:52
good. Something feels good about it. It's well done.
1:59:54
It's like, oh, okay, I never thought of it this way.
1:59:57
This just feels good. Sometimes the outrage
1:59:59
mobs, even if-
1:59:59
if they have a good point behind it, when they're like
2:00:02
mocking and derisive and just aggressive,
2:00:04
you're with us or against us, this
2:00:07
fucking- This is why I delete my tweets. Yeah,
2:00:10
why'd you do that?
2:00:12
I was, you know, I missed your tweets.
2:00:14
You know what it is? The algorithm promotes
2:00:17
toxicity. Yeah.
2:00:19
And like, you know, I think
2:00:22
Elon has a much better chance of fixing it than the previous
2:00:25
regime. Yeah.
2:00:28
But to solve this problem, to
2:00:30
build a social network that is actually not
2:00:32
toxic without
2:00:35
moderation. Like
2:00:39
not to stick but care. So like where people look
2:00:43
for goodness, to
2:00:45
make it catalyze the process of connecting
2:00:47
cool people and being cool to each other. Yeah.
2:00:51
Without ever censoring. Without ever censoring.
2:00:53
And like Scott Alexander has a
2:00:55
blog post I like where he talks about like moderation is not censorship,
2:00:58
right? Like all moderation you
2:01:00
want to put on Twitter,
2:01:01
right? Like you could totally make this
2:01:03
moderation
2:01:04
like just a,
2:01:06
you don't have to block it for everybody. You
2:01:08
can just have like a filter button,
2:01:10
right? That people can turn off if they were like safe search for Twitter,
2:01:12
right? Like someone could just turn that off, right? So
2:01:14
like, but then you'd like take this idea to an extreme, right?
2:01:17
Well,
2:01:17
the network should just show you, this
2:01:20
is a couch surfing CEO thing, right? If
2:01:22
it shows you right now these algorithms are
2:01:24
designed to maximize engagement. Well,
2:01:26
it turns out Outrage maximizes engagement.
2:01:28
Quirk of human, quirk of the human
2:01:30
mind, right?
2:01:31
Just this, I fall for it, everyone falls for it. So
2:01:35
yeah, you got to figure out how to maximize for something other
2:01:37
than engagement.
2:01:38
And I actually believe that you can make money with
2:01:40
that too. So it's not, I don't think engagement
2:01:42
is the only way to make money. I actually think it's incredible
2:01:45
that we're starting to see, I think again,
2:01:47
Yolen's doing so much stuff right with Twitter like charging
2:01:49
people money.
2:01:50
As soon as you charge people money, they're no longer
2:01:52
the product, they're the customer.
2:01:55
And then they can start building something that's good
2:01:57
for the customer and not good for the other customer,
2:01:59
which is the ad agency. As in hasn't
2:02:01
picked up steam I Pay
2:02:04
for Twitter doesn't even get me anything. It's my donation
2:02:06
to this new business model. Hopefully working out
2:02:08
sure But you know you for this business
2:02:10
model to work. It's like most people
2:02:13
should be signed up to Twitter and so the
2:02:15
way was There
2:02:17
was something perhaps not compelling or something
2:02:19
like this to people think you need most people
2:02:21
at all I think that why do I need most
2:02:23
people right? I don't make an 8,000 person company
2:02:26
make a 50 person company
2:02:28
Well, so speaking of which
2:02:32
You worked at Twitter for a bit I did as
2:02:35
an intern The
2:02:37
world's greatest intern. Yeah. All right,
2:02:40
there's been better. That's been better Tell
2:02:43
me about your time at Twitter. How did it come about
2:02:46
and what did you learn from the experience? So
2:02:48
I deleted
2:02:51
my first Twitter in 2010 I had
2:02:54
over hundred thousand followers
2:02:56
back when that actually meant something and I
2:03:00
Just saw you know
2:03:03
My co-worker summarized it well. He's
2:03:05
like
2:03:06
whenever I see someone's Twitter page
2:03:08
I either think the same of them or less
2:03:10
of them. I never think more of them Yeah,
2:03:13
right like like, you know, I don't want to mention any
2:03:15
names but like some people who like, you know, maybe you would
2:03:17
like read their books and you would respect them you see
2:03:19
them on Twitter and You're like Okay,
2:03:22
dude But
2:03:25
there's some people with same
2:03:27
You know who I respect a lot are
2:03:29
people that just post really good technical stuff.
2:03:32
Yeah, and I
2:03:34
guess I Don't
2:03:36
know. I think I respect them more for it because
2:03:38
you realize oh this wasn't There's
2:03:41
like so much depth to
2:03:43
this person to their technical understanding of so many different
2:03:46
topics
2:03:46
Okay, so I try to follow people. I
2:03:49
try to consume stuff. That's technical
2:03:52
machine learning content
2:03:53
there's probably a few
2:03:55
of those people and The
2:03:58
problem is inherently what?
2:03:59
the algorithm rewards, right? And
2:04:02
people think about these algorithms, people think that they
2:04:04
are terrible, awful things. And you know, I love that Elon
2:04:06
open sourced it. Because I mean, what it
2:04:09
does is actually pretty obvious. It just predicts
2:04:11
what you are likely to retweet and like, and
2:04:14
linger on.
2:04:15
So what all these algorithms do, so what TikTok does, so
2:04:17
all these recommendation engines do.
2:04:18
And it turns
2:04:21
out that the thing that you are most likely
2:04:23
to interact with is outrage.
2:04:25
And that's a quirk of the human condition. I
2:04:30
mean, and there's different flavors of outrage. It doesn't have
2:04:32
to be, it could be mockery. You
2:04:35
could be outraged. The topic of outrage could be different.
2:04:37
It could be an idea. It could be a person. It could be, and maybe
2:04:41
there's a better word than outrage. It could be drama. Sure.
2:04:44
Drama. All this kind of stuff. Yeah. But it doesn't
2:04:46
feel like when you consume it, it's a constructive
2:04:48
thing for the individuals that consume it in the
2:04:51
long term.
2:04:51
Yeah. So my time there,
2:04:54
I absolutely couldn't believe, you know,
2:04:56
I got crazy amount
2:04:58
of hate, you know, just on
2:05:00
Twitter for working at Twitter. It seemed like people
2:05:03
associated with this, I think maybe you were
2:05:06
exposed to some of this. So connection to Elon
2:05:08
or is it working at Twitter?
2:05:09
Twitter and Elon, like the whole...
2:05:12
Because Elon's gotten a bit spicy during
2:05:14
that time. A bit political,
2:05:16
a bit... Yeah. Yeah.
2:05:18
You know, I remember one of my tweets, it was never go
2:05:20
full Republican, and Elon liked it. You
2:05:23
know, I think... Oh
2:05:29
boy. Yeah, I mean, there's
2:05:31
a roller coaster of that, but being political on
2:05:33
Twitter,
2:05:34
boy. Yeah. And
2:05:37
also being, just attacking
2:05:39
anybody on Twitter, it comes back at
2:05:41
you harder. And if
2:05:43
his political end attacks. Sure. Sure,
2:05:46
absolutely. And then letting
2:05:50
sort of de-platform
2:05:53
people back on, even
2:05:56
adds more fun to the
2:05:58
beautiful chaos. I was hoping, and
2:06:01
I remember when Elon talked about buying Twitter
2:06:05
six months earlier, he was talking about
2:06:07
a principled commitment to free
2:06:09
speech.
2:06:10
I'm a big believer and
2:06:12
fan of that. I would love to see an actual
2:06:15
principled commitment to free speech. Of
2:06:18
course, this isn't quite what happened.
2:06:20
Instead of the oligarchy deciding
2:06:22
what to ban,
2:06:23
you had a monarchy deciding what to ban.
2:06:26
Instead of all the Twitter files,
2:06:28
shadow, really, the oligarchy
2:06:31
just decides what. Cloth masks are ineffective
2:06:33
against COVID. That's a true statement. Every doctor
2:06:35
in 2019 knew it and now I'm banned on Twitter for saying
2:06:38
it. Interesting. Oligarchy.
2:06:40
Now you have a monarchy and
2:06:42
he bans things he doesn't like. It's
2:06:46
just different power and
2:06:49
maybe I align more with him than with the oligarchy. But
2:06:52
it's not free speech. I
2:06:55
feel like
2:06:56
being a free speech absolutist on a social network
2:06:58
requires you to also have tools for
2:07:01
the individuals to
2:07:04
control what they consume easier.
2:07:08
Not censor, but just
2:07:10
control, oh, I'd like to see more cats
2:07:13
and less politics. This
2:07:15
isn't even remotely controversial. This is just saying
2:07:17
you want to give paying customers for a product what they want.
2:07:21
Not through the process of censorship, but through the process
2:07:23
of like- It's individualized, right? It's
2:07:25
individualized transparent censorship, which is honestly
2:07:27
what I want. What is an ad blocker? It's individualized
2:07:29
transparent censorship, right? Yeah, but censorship
2:07:32
is a strong word and
2:07:34
people are very sensitive too. I know, but
2:07:37
I just use words to describe what they functionally are and
2:07:39
what is an ad blocker. It's just censorship. But
2:07:41
I love what you're censoring. I'm
2:07:43
looking at you, I'm censoring
2:07:46
everything else out when my mind is focused
2:07:48
on you. You can use the word censorship
2:07:51
that way, but usually when people get very sensitive
2:07:53
about the censorship thing, I
2:07:55
think when anyone is allowed to
2:07:57
say anything, you should probably
2:07:59
have-
2:07:59
tools that maximize
2:08:03
the quality of the experience for individuals. So,
2:08:05
you know, for me, like what I really
2:08:07
value, boy, it would be amazing
2:08:09
to somehow figure out how to do that.
2:08:12
I love disagreement and debate and
2:08:14
people who disagree with each other
2:08:16
disagree with me, especially in the space of ideas, but
2:08:19
the high quality ones. So not derision,
2:08:21
right? Maslow's hierarchy of argument.
2:08:24
I think it's a real word for it. Probably.
2:08:26
There's just the way of talking that's like snarky
2:08:28
and so on that somehow is gets
2:08:31
people on Twitter and they get excited and so on.
2:08:33
You have like ad hominem refuting the central point.
2:08:35
I've like seen this as an actual pyramid. Yeah, it's yeah.
2:08:38
And it's like all of it,
2:08:40
all the wrong stuff is attractive to people.
2:08:42
I mean, we can just train a classifier to absolutely say what
2:08:44
level of Maslow's hierarchy of argument
2:08:47
are you at? And if it's ad hominem, like, okay,
2:08:49
cool. I turned on the no ad hominem filter.
2:08:52
I wonder if there's a social
2:08:54
network that will allow you to have that kind of filter.
2:08:56
Yeah, so here's
2:08:59
a problem with that. It's
2:09:01
not going to win in a free market.
2:09:04
What wins in a free market is all
2:09:06
television today is reality television because it's engaging.
2:09:08
If engaging
2:09:11
is what wins in a free market, right? So
2:09:13
it becomes hard to keep these other more
2:09:15
nuanced values.
2:09:16
Well,
2:09:19
okay, so that's the experience of being on Twitter,
2:09:22
but then you got a chance to also
2:09:24
and together with other engineers and
2:09:26
with Elon sort of look brainstorm
2:09:28
when you step into a code base has been around
2:09:31
for a long time. There's other social
2:09:33
networks, Facebook, this is old
2:09:35
code bases. And you step in and see,
2:09:37
okay, how do we make
2:09:40
with a fresh mind progress
2:09:42
on this code base? Like what did you learn about
2:09:44
software engineering, about programming from just experiencing
2:09:47
that? So my
2:09:49
technical recommendation to Elon, and I said
2:09:51
this on the Twitter spaces afterward, I said this
2:09:54
many times during my brief internship,
2:09:58
was that, you need refactors
2:10:01
before features. This
2:10:03
code base was, and look,
2:10:06
I've worked at Google, I've worked at Facebook. Facebook
2:10:08
has the best code, then
2:10:10
Google, then Twitter. And
2:10:12
you know what? You can know this because look at
2:10:14
the machine learning frameworks, right? Facebook released PyTorch,
2:10:17
Google released TensorFlow and Twitter released, okay,
2:10:21
so you know. It's a proxy,
2:10:24
but yeah, the Google code base
2:10:26
is quite interesting. There's a lot of really good software engineers
2:10:28
there, but the code base is very large. The
2:10:30
code base was good in 2005, right? It
2:10:33
looks like 2005. There's so many products, so many
2:10:35
teams, right? It's very difficult to, I
2:10:38
feel like Twitter does less, obviously
2:10:42
much less than Google
2:10:44
in terms of like the set of features,
2:10:48
right? So like it's, I
2:10:50
can imagine the number of software
2:10:52
engineers that could recreate Twitter
2:10:54
is much smaller than to recreate Google. Yeah,
2:10:56
I still believe in the amount of hate
2:10:59
I got for saying this, that 50 people
2:11:01
could build and maintain Twitter pretty
2:11:03
comfortably. What's the nature of the hate? But
2:11:07
you don't know what you're talking about. You know what it is? And
2:11:10
it's the same, this is my summary of like the hate I get
2:11:12
on Hacker News. It's like,
2:11:14
when I say I'm going to do something, they
2:11:17
have
2:11:19
to believe that it's impossible. Because
2:11:23
if doing things was possible, they'd
2:11:25
have to do some soul searching and ask the question,
2:11:27
why didn't they do anything? So when you say, and
2:11:30
I do think that's where the hate comes from. When you say, well,
2:11:32
there's a core truth to that. Yeah, so when you say I'm going
2:11:34
to solve self-driving,
2:11:37
people go like, what are your credentials? What
2:11:40
the hell are you talking about? This is an extremely
2:11:42
difficult problem. Of course, you're a noob that doesn't understand
2:11:44
the problem deeply. I
2:11:47
mean, that was the same nature of hate
2:11:49
that probably Elon got when he first talked about autonomous
2:11:51
driving.
2:11:53
But there's pros
2:11:55
and cons to that. Because there is experts
2:11:57
in this world. No, but
2:11:59
the market. aren't experts. The people
2:12:02
who are mocking are not experts with carefully
2:12:04
reasoned arguments about why you need 8,000 people
2:12:06
to run a bird app. But the people
2:12:09
are going to lose their jobs.
2:12:12
Well, that, but also there's the software
2:12:14
engineers that probably could have said, no, it's a lot more complicated
2:12:16
than you realize, but maybe it doesn't need to be so
2:12:18
complicated. You know,
2:12:20
some people in the world like to create complexity.
2:12:22
Some people in the world thrive under complexity, like lawyers,
2:12:25
right? Lawyers want the world to be more complex because you
2:12:27
need more lawyers and you need more legal hours, right? I
2:12:29
think that's another. If
2:12:31
there's two great evils in the world, it's centralization
2:12:34
and complexity. Yeah. And the
2:12:36
one of the sort of hidden side effects
2:12:40
of software engineering is
2:12:44
like finding pleasure and complexity.
2:12:47
I mean, I don't remember just taking
2:12:50
all the software engineering courses and just doing programming
2:12:52
and this is just coming up in this
2:12:56
object-oriented programming kind of idea.
2:13:00
Not often do people tell you, do
2:13:02
the simplest possible thing. A professor,
2:13:06
a teacher is not going to get in front
2:13:08
like, this is the simplest way to do
2:13:10
it. They'll say, this is the
2:13:12
right way and the right way, at least for
2:13:15
a long time, especially I came
2:13:17
up with like Java, right? There's
2:13:20
so much boilerplate, so much like, so
2:13:23
many classes, so many like designs
2:13:26
and architectures and so on, like planning for
2:13:29
features far into the future
2:13:31
and planning poorly and all this
2:13:33
kind of stuff. And then there's this like code
2:13:35
base that follows you along and puts pressure on
2:13:37
you and nobody knows
2:13:39
what like parts, different parts do, which
2:13:42
slows everything down. There's a kind of bureaucracy that's
2:13:44
instilled in the code as a result of that, but
2:13:46
then you feel like, oh well I follow
2:13:49
good software engineering practices. It's an
2:13:51
interesting trade-off because then you look at like the
2:13:53
ghetto-ness of like Pearl and
2:13:56
the old like, how quickly you just
2:13:58
write a couple lines and just get stuff done.
2:13:59
that trade-off is interesting or bash
2:14:02
or whatever, these kind of ghetto things you can do
2:14:04
on Linux. One of my favorite things
2:14:06
to look at today is how much do you trust your tests?
2:14:09
We've put a ton of effort in comma and I put a ton
2:14:11
of effort in tiny grad into making sure,
2:14:14
if you change the code and the tests pass,
2:14:17
that you didn't break the code. Now, obviously,
2:14:19
it's not always true. But the
2:14:21
closer that is to true, the more you trust your
2:14:23
tests, the more you're like, oh, I got a pull request and
2:14:25
the tests pass, I feel okay to merge
2:14:27
that, the faster you can make progress. You're always programming
2:14:30
your tests in mind, developing tests with
2:14:32
that in mind that if it passes, it should be good.
2:14:34
Twitter had a- Not that.
2:14:37
It was impossible to make
2:14:39
progress in the code base. What other
2:14:41
stuff can you say about the code base that made it difficult?
2:14:45
What are some interesting quirks broadly
2:14:47
speaking from that compared
2:14:50
to just your experience with comma and
2:14:52
everywhere else? The real thing that
2:14:55
I spoke to a bunch of,
2:14:59
individual contributors at Twitter and I just
2:15:01
had a test. I'm like, okay, so
2:15:03
what's wrong with this place? Why does this code look like this? They
2:15:06
explained to me what Twitter's promotion system
2:15:08
was. The way that you got promoted
2:15:10
to Twitter was you wrote a library that
2:15:12
a lot of people used. Some
2:15:17
guy wrote an NGINX replacement
2:15:19
for Twitter. Why does Twitter need an NGINX
2:15:21
replacement? What was wrong with NGINX? You
2:15:24
see, you're not going to get promoted if you use
2:15:26
NGINX. But if you write a replacement
2:15:29
and lots of people start using it as the Twitter
2:15:31
front-end for their product, then you're going to get promoted.
2:15:34
So interesting because from an individual
2:15:36
perspective, how do you incentivize,
2:15:39
how do you create the kind of incentives that will
2:15:41
lead to a great code base? Okay,
2:15:44
what's the answer to that? So
2:15:47
what I do at comma and at
2:15:52
TinyCorp is you have to explain it to me. You have to
2:15:54
explain to me what this code does. And
2:15:56
if I can sit there and come up with a simpler
2:15:58
way to do it, you have to... You have
2:16:01
to agree with me about the simpler way. Obviously,
2:16:03
we can have a conversation about this. It's not
2:16:05
dictatorial, but if you're like, wow, wait,
2:16:07
that actually is way simpler.
2:16:10
Like the simplicity is important.
2:16:12
Right? But that requires people
2:16:14
that overlook the code at the
2:16:17
highest levels to be like, okay.
2:16:19
It requires technical leadership, you trust. Yeah,
2:16:22
technical leadership. So
2:16:24
managers or whatever should have to have
2:16:26
technical savvy, deep technical savvy. Managers
2:16:29
should be better programmers than the people who they manage.
2:16:32
Yeah. And that's not always obvious
2:16:35
to create, especially large companies. Managers
2:16:37
get soft. And like, you know, and this is just, I've instilled
2:16:40
this culture at Kama and Kama has better programmers
2:16:42
than me who work there. But you know, again,
2:16:45
I'm like the, you know, the old guy from Good Will Hunting. It's
2:16:47
like, look, man,
2:16:48
you know, I might not be as
2:16:50
good as you, but I can see the difference between me and you. Right?
2:16:53
And like, this is what you need. This is what you need at the top. Or
2:16:55
you don't necessarily need the manager to be the absolute
2:16:58
best. I shouldn't say that, but like they need
2:17:00
to be able to recognize skill. Yeah.
2:17:02
And have good intuition, intuition
2:17:05
that's laden with wisdom from all the
2:17:07
battles of trying to reduce complexity
2:17:09
and code bases. You know, I took a, I took a political
2:17:12
approach at Kama too, that I think is pretty interesting. I think Elon
2:17:14
takes the same political approach. You
2:17:16
know, Google had no politics
2:17:19
and what ended up happening is the absolute worst kind of politics
2:17:21
took over.
2:17:22
Kama has an extreme amount of politics and they're
2:17:25
all mine and no dissidents is tolerated.
2:17:28
So it's a dictatorship. Yep. It's
2:17:30
an absolute dictatorship. Right. Elon does
2:17:32
the same thing. Now, the thing about my dictatorship is
2:17:34
here are my values. Yeah. So
2:17:37
it's transparent. It's transparent. It's
2:17:39
a transparent dictatorship. Right. And you can
2:17:41
choose to opt in or, you know, you get free exit, right? That's the beauty of companies.
2:17:44
If you don't like the dictatorship, you quit.
2:17:46
So you
2:17:48
mentioned rewrite before
2:17:50
or refactor before features.
2:17:54
If you were to refactor the Twitter code base,
2:17:56
what would that look like? And maybe also
2:17:58
comment on how difficult the code is. is it to refactor?
2:18:01
The main thing I would do is first of all,
2:18:03
identify the pieces and then put tests
2:18:05
in between the pieces,
2:18:07
right? So there's all these different Twitter as a microservice
2:18:09
architecture, um, all
2:18:12
these different microservices. And the
2:18:14
thing that I was working on there, look, like, you know, George
2:18:18
didn't know any JavaScript, he asked how to fix
2:18:19
search, blah, blah, blah, blah. Look,
2:18:21
man, like the thing is
2:18:24
like, I just, you know, I'm upset that the
2:18:26
way that this whole thing was portrayed, because
2:18:28
it wasn't like, it wasn't like taken by people,
2:18:30
like, honestly, it wasn't like, it
2:18:32
was taken by people who started out
2:18:34
with a bad faith assumption. Yeah. And
2:18:36
I mean, I look, I can't like, and you as a programmer,
2:18:39
just being transparent out there, actually having
2:18:41
like fun and like, this is what programmers
2:18:44
should be about. It's just like, I love that Elon gave
2:18:46
me this opportunity. Yeah. Like really it does.
2:18:48
And like, you know, he came up with my, the day I quit, he came
2:18:50
up with my Twitter spaces afterward and we had a conversation
2:18:53
like, I just, I respect that so much.
2:18:55
Yeah. And it's also inspiring to just engineers
2:18:57
and programmers and just, it's cool. It should
2:18:59
be fun. The people that were hating on it is
2:19:01
like, oh man. It
2:19:03
was fun. It was fun. It was stressful,
2:19:06
but I felt like, you know, it was not like a cool like
2:19:08
point in history and like, I hope I was useful and
2:19:10
probably kind of wasn't, but like, maybe
2:19:12
I'm also were one of the people that
2:19:15
kind of made a strong case to refactor. Yeah.
2:19:17
And that that's a really
2:19:19
interesting thing to raise. Like
2:19:21
maybe that is the right, you know, the
2:19:24
timing of that is really interesting. If you look at just the development
2:19:26
of autopilot, you know, going from
2:19:29
mobile eye to just like more,
2:19:32
if you look at the history of semi-autonomous
2:19:34
driving in Tesla is, is
2:19:36
more and more
2:19:38
like you could say refactoring or,
2:19:41
or starting from scratch, redeveloping from scratch. It's
2:19:43
refactoring all the way down. And like,
2:19:46
and the question is like, can you do that sooner? Can
2:19:49
you maintain product profitability?
2:19:52
And like, what's the, what's the right time to do it? How
2:19:55
do you do it? You know, on any one day,
2:19:57
it's like, you don't want to pull off the band-aids. Like
2:19:59
it's.
2:19:59
Like everything works
2:20:02
is just like little fixed here and there,
2:20:04
but maybe starting from scratch. This
2:20:06
is the main philosophy of TinyGrad. You have never
2:20:09
refactored enough. Your code can get smaller,
2:20:11
your code can get simpler, your ideas can be more
2:20:13
elegant.
2:20:14
But would you consider,
2:20:16
you know, say you were like running
2:20:19
Twitter development teams, engineering
2:20:21
teams, would
2:20:23
you go as far as like different programming language?
2:20:26
Just go that far. I
2:20:28
mean, the first thing that I would do
2:20:31
is build tests. The first thing I would
2:20:33
do is get a CI to
2:20:36
where people can trust to make changes.
2:20:39
So that if you keep- Before I touched any
2:20:41
code, I would actually say, no one touches
2:20:44
any code. The first thing we do is we test this code
2:20:46
base. I mean, this is classic, this is how you approach a legacy code
2:20:48
base. This is like what any, how to approach
2:20:50
a legacy code base book will tell you.
2:20:52
So, and then you hope
2:20:54
that there's modules that can
2:20:57
live on for a while. And
2:20:59
then you add new ones, maybe in a different
2:21:01
language or- Before we add
2:21:03
new ones, we replace old ones. Yeah, meaning
2:21:06
like replace old ones with something simpler. We
2:21:08
look at this like this thing that's 100,000
2:21:10
lines and we're like, well, okay,
2:21:12
maybe this did even make sense in 2010, but
2:21:14
now we can replace this with an open source thing.
2:21:17
Right? Yeah. And you know, we
2:21:19
look at this here, here's another 50,000 lines. Well,
2:21:21
actually, you know, we can replace this with 300 lines of go. And
2:21:25
you know what? We trust that the go actually replaces
2:21:27
this thing because all the tests still pass. So
2:21:29
step one is testing. Yeah. And
2:21:31
then step two is like the programming languages and afterthought,
2:21:34
right? You know, let a whole lot of people compete, be like,
2:21:36
okay, who wants to rewrite a module, whatever language you want to
2:21:38
write it in, just the tests have to pass. And
2:21:41
if you figure out how to make the test pass, but
2:21:43
break the site, that's, we got to go back to step
2:21:45
one. Step one is get tests that
2:21:47
you trust in order to make changes in the code base.
2:21:49
I wonder how hard it is to, because I'm with you
2:21:51
on testing and everything. Hey, you have from
2:21:54
tests to like asserts to everything, code
2:21:57
is just covered in this because.
2:21:59
it should be very
2:22:02
easy to make rapid changes
2:22:05
and no, that's not gonna break everything. And
2:22:08
that's the way to do it. But I wonder how difficult
2:22:10
is it to
2:22:11
integrate tests into
2:22:14
a code base that doesn't have many of them. So I'll
2:22:16
tell you what my plan was at Twitter. It's actually similar to
2:22:18
something we use at Comma. So at Comma we have this thing called process
2:22:20
replay. And we have a bunch of routes
2:22:22
that'll be run through. So Comma's a microservice architecture
2:22:25
too. We have microservices in the driving.
2:22:27
Like we have one for the cameras, one for the sensor, one for the
2:22:29
planner, one for
2:22:31
the model. And we
2:22:34
have an API, which the microservices talk
2:22:36
to each other with. We use this custom thing called Serial,
2:22:38
which uses a ZMQ. Twitter uses
2:22:42
Thrift. And then it uses
2:22:45
this thing called Finagle, which is a Scala RPC backend.
2:22:50
But this doesn't even really matter. The Thrift and Finagle
2:22:53
layer was a great place, I
2:22:56
thought, to write tests. To start building
2:22:58
something that looks like process replay. So Twitter
2:23:00
had some stuff that looked kind of like
2:23:02
this, but it wasn't offline. It
2:23:05
was only online. So you could ship
2:23:07
a modified version of it, and
2:23:09
then you could redirect
2:23:11
some of the traffic to your modified version and diff those
2:23:13
too. But it was all online. There
2:23:15
was no CI in the traditional sense.
2:23:17
I mean, there was some, but it was not full coverage. So
2:23:19
you can't run all of Twitter offline
2:23:22
to test something. Then this was another problem. You can't
2:23:24
run all of Twitter.
2:23:25
Period. Anyone person
2:23:27
can't run. Twitter runs in three data
2:23:29
centers, and that's it. There's no other
2:23:32
place you can run Twitter, which is like,
2:23:34
George, you don't understand. This is modern
2:23:36
software development. No, this is bullshit. Why
2:23:39
can't it run on my laptop?
2:23:41
What are you doing? Twitter can run on, yeah, okay.
2:23:43
Well, I'm not saying you're gonna download the whole
2:23:45
database to your laptop, but I'm saying all the middleware
2:23:47
and the front end should run on my laptop, right? That
2:23:50
sounds really compelling. Yeah. But
2:23:53
can that be achieved by
2:23:56
a code base that grows over the years? I
2:23:58
mean, the three data centers. doesn't have to be, right? Because
2:24:01
they're totally different designs. The
2:24:03
problem is more like,
2:24:05
why did the code base have to grow?
2:24:07
What new functionality has been added to
2:24:09
compensate for the lines
2:24:11
of code that are there?
2:24:13
One of the ways to explain is that the
2:24:15
incentive for software developers to move up in the company
2:24:18
is to add code. To
2:24:20
add, especially large. And you know what? The incentive for
2:24:22
politicians to move up in the political structure is to add laws.
2:24:25
Yeah. Same problem. Yeah.
2:24:28
Yeah. The flip
2:24:30
side is to simplify, simplify, simplify. I
2:24:33
mean, you know what? This is something that
2:24:35
I do differently from Elon with
2:24:37
comma about self-driving cars.
2:24:40
You know, I hear the new version's gonna come out
2:24:42
and the new version is not gonna be better, but
2:24:45
at first, and it's gonna require a ton of refactors.
2:24:48
I say, okay, take as long as you need. Like,
2:24:51
you convinced me this architecture's better? Okay, we
2:24:53
have to move to it. Even if it's not gonna
2:24:55
make the product better tomorrow, the top
2:24:57
priority is getting the architecture right.
2:25:00
So what do you think about
2:25:01
sort of a thing where
2:25:03
the product is online?
2:25:05
So I guess,
2:25:07
would you do a refactor? If you ran
2:25:10
engineering on Twitter, would you just do a refactor?
2:25:12
How long would it take? What would that mean for the
2:25:14
running of the actual service?
2:25:17
You know, and...
2:25:21
I'm not the right person to run Twitter.
2:25:23
I'm just not. And that's the problem. Like,
2:25:26
I don't really know. I don't really know if
2:25:28
that's... A common thing that I thought
2:25:30
a lot while I was there was whenever I thought something
2:25:32
that was different to what Elon thought,
2:25:34
I'd have to run something in the back of my head reminding
2:25:37
myself
2:25:38
that Elon is the richest man in
2:25:40
the world. And in
2:25:42
general, his ideas are better than mine.
2:25:45
Now, there's a few things I think I do understand
2:25:48
and know more about, but
2:25:50
like in general,
2:25:52
I'm not qualified to run Twitter. I'm not necessarily
2:25:55
qualified, but like, I don't think I'd be that good at it. I
2:25:57
don't think I'd be good at it.
2:25:58
I don't think I'd really be good at running.
2:25:59
an engineering organization at scale.
2:26:02
I think I could
2:26:04
lead a very good refactor
2:26:07
of Twitter,
2:26:08
and it would take like six months to a year. And
2:26:11
the results to show at the end of it would be
2:26:13
feature development in general takes
2:26:16
10x less time, 10x less man hours.
2:26:18
That's what I think I could actually do. Do
2:26:21
I think that it's the right decision for the business
2:26:24
above my pay grade?
2:26:26
Yeah,
2:26:28
but a lot of these kinds of decisions are
2:26:30
above everybody's pay grade. I don't want to be a manager.
2:26:33
I don't want to do that. Like if
2:26:35
you really forced me to, yeah, it would make me
2:26:37
maybe
2:26:40
make me upset if I had to make
2:26:42
those decisions. I don't want to. Yeah,
2:26:46
but a refactor is so compelling.
2:26:49
If this is to become something
2:26:51
much bigger than what Twitter was, it
2:26:54
feels like a refactor
2:26:56
has to be coming at some point. George, you're
2:26:58
a junior software engineer. Every junior software
2:27:01
engineer wants to come in and refactor the whole code.
2:27:04
OK, that's like your opinion,
2:27:06
man. Yeah,
2:27:09
sometimes they're right. Well,
2:27:11
whether they're right or not, it's definitely not
2:27:13
for that reason, right? It's definitely not a question of engineering
2:27:16
prowess. It is a question of maybe what the priorities are for the
2:27:18
company. And I did get more intelligent
2:27:21
feedback from people, I think, in good faith, like saying
2:27:23
that. Actually,
2:27:26
from Elon.
2:27:26
And from Elon, people
2:27:29
were like, well, a stop the world refactor
2:27:32
might be great for engineering, but you know we have a business
2:27:34
to run.
2:27:35
And hey, above
2:27:37
my pay grade. What did you think about Elon
2:27:39
as an engineering leader, having
2:27:42
to experience him in the most chaotic
2:27:44
of spaces, I would say?
2:27:51
My respect for him is unchanged. And I did have
2:27:53
to think a lot more deeply about some
2:27:55
of the decisions he's forced to make. About.
2:27:59
the tensions within those,
2:28:01
the trade-offs within those decisions?
2:28:05
About like a whole like
2:28:08
matrix coming at him. I think that's Andrew
2:28:10
Tate's word for it. Sorry to borrow it. Also,
2:28:12
bigger than engineering, just everything. Yeah,
2:28:16
like
2:28:17
the war on the woke. Yeah.
2:28:20
Like, it just, man,
2:28:23
and like, he doesn't
2:28:25
have to do this, you know? He doesn't
2:28:27
have to. So, like, parag
2:28:29
and go chill at the Four Seasons of Maui, you
2:28:32
know? But
2:28:33
see, one person I respect and one person I don't.
2:28:36
So his heart is in the right place
2:28:38
fighting, in this case, for this ideal
2:28:40
of the freedom of expression.
2:28:43
Well, I wouldn't define the ideal so simply.
2:28:45
I think you can define the ideal
2:28:47
no more than just saying
2:28:50
Elon's idea of a good world.
2:28:52
Freedom of expression is...
2:28:54
To you, it's still, the downside
2:28:57
of that is the monarchy.
2:28:58
Yeah, I mean, monarchy has
2:29:01
problems, right? But I mean, would
2:29:04
I trade right now the
2:29:06
current oligarchy, which runs America
2:29:08
for the monarchy? Yeah, I would. Sure.
2:29:11
For the Elon monarchy? Yeah, you know why? Because
2:29:13
power would cost one cent a kilowatt hour.
2:29:16
A tenth of a cent a kilowatt hour.
2:29:18
What do you mean? Right now, I pay
2:29:20
about 20 cents a kilowatt hour for electricity
2:29:23
in San Diego. It's like the
2:29:25
same price you paid in 1980. What
2:29:27
the hell? So you would see a lot of
2:29:29
innovation with Elon.
2:29:31
Maybe it'd have some hyper loops.
2:29:33
Yeah. Right? And I'm willing
2:29:35
to make that trade off, right? I'm willing to make... And this is
2:29:37
why. You know, people think that like dictators take
2:29:39
power through some untoward
2:29:41
mechanism. Sometimes they do, but usually it's
2:29:44
because the people want them. And
2:29:46
the downsides of a dictatorship,
2:29:48
I feel like we've gotten to a point now with the oligarchy where,
2:29:51
yeah, I would prefer the dictator.
2:29:53
What
2:29:56
do you think about Scala as a programming language?
2:30:01
I liked it more than I thought. I did the tutorials.
2:30:03
Like I was very new to it. Like it would take me six months to be able to
2:30:05
write like good Scala.
2:30:07
I mean, what did you learn about learning a new programming language
2:30:09
from that? I love doing
2:30:11
like new programming, I did tutorials and doing them. I did all this
2:30:13
for Rust. It
2:30:17
keeps some of its upsetting JVM roots,
2:30:20
but it is a much nicer. In
2:30:22
fact, I almost don't know why Kotlin took off
2:30:24
and not Scala.
2:30:26
I think Scala has some beauty that Kotlin lacked.
2:30:30
Whereas Kotlin felt a lot more.
2:30:33
I mean, it was almost like, I don't know if it actually was
2:30:35
a response to Swift, but that's kind of what it felt like.
2:30:38
Like Kotlin looks more like Swift and Scala looks more
2:30:40
like, well, I could have a functional programming language,
2:30:42
more like an OCaml or Haskell. Let's
2:30:44
actually just explore, we touched it a little bit,
2:30:46
but just on the art, the
2:30:49
science and the art of programming. For
2:30:51
you personally, how much of your programming is done with
2:30:53
GPT currently? None.
2:30:55
I don't use it at all.
2:30:57
Because you prioritize simplicity so much.
2:30:59
Yeah, I find that a
2:31:01
lot of it is noise. I do use
2:31:04
VS code
2:31:05
and I do like
2:31:07
some amount of autocomplete.
2:31:09
I do like a
2:31:10
very, feels like rules-based autocomplete.
2:31:13
An autocomplete is going to complete the variable name
2:31:15
for me, so I'm going to type it, I can just press tab. That's
2:31:17
nice, but I don't want it autocomplete. You know what
2:31:19
I hate? When autocompletes, when I type the word for
2:31:22
and it puts like two parentheses
2:31:24
and two semicolons and two braces, I'm like, oh man.
2:31:28
Well, let me with VS code
2:31:31
and GPT with codecs,
2:31:35
you can kind of brainstorm. I
2:31:37
find, I'm like
2:31:40
probably the same as you, but I like
2:31:42
that it generates code and you basically
2:31:45
disagree with it and write something simpler. But
2:31:47
to me, that somehow is like inspiring,
2:31:50
it makes me feel good. It also gamifies the simplification
2:31:52
process because I'm like, oh yeah, you
2:31:55
dumb AI system. You think this is the way
2:31:57
to do it. I have a simpler thing here. It just
2:31:59
constantly.
2:31:59
reminds me of bad stuff. I
2:32:02
mean, I tried the same thing with rap, right? I tried the same
2:32:04
thing with rap, and I actually think of a much better program than rapper,
2:32:07
but I even tried, I was like, okay, can we get some inspiration
2:32:09
from these things for some rap lyrics? And
2:32:12
I just found that it would go back to the most cringy
2:32:15
tropes and dumb rhyme schemes,
2:32:17
and I'm like, yeah, this is what the code looks
2:32:19
like too. I think you and
2:32:21
I probably have different thresholds for cringe code. You
2:32:24
probably hate cringe code. So
2:32:26
it's for you.
2:32:30
Boilerplate is a part of code. Some
2:32:35
of it, yeah,
2:32:39
and some of it is just faster lookup. Because
2:32:43
I don't know about you, but I don't remember everything. I'm
2:32:46
offloading so much of my memory about
2:32:50
different functions, library functions, all that kind
2:32:52
of stuff. GPT
2:32:55
just is very fast. It's standard
2:32:57
stuff, standard library
2:33:00
stuff, basic stuff that everybody uses.
2:33:03
I think that, I
2:33:08
don't know, I mean, there's just a little of this in Python.
2:33:10
Maybe if I was
2:33:12
coding more in other languages, I would consider it
2:33:14
more, but I feel like Python already
2:33:17
does such a good job of removing
2:33:19
any Boilerplate.
2:33:20
That's true. It's the closest thing you can get to pseudocode,
2:33:23
right? Yeah, that's true. That's
2:33:25
true. I'm like,
2:33:27
yeah, sure. If I like, yeah, I'm great, GPT. Thanks
2:33:29
for reminding me to free my variables. Unfortunately,
2:33:32
you didn't really recognize the scope correctly
2:33:34
and you can't free that one, but you put
2:33:36
the freeze there and I get it. Fiber.
2:33:41
Whenever I've used fiber for certain things,
2:33:43
like design or whatever, it's
2:33:45
always, you come back. I think that's probably
2:33:47
closer, my experience with fiber is closer to your
2:33:50
experience with programming
2:33:50
with GPT. You're just frustrated
2:33:53
and feel worse about the whole process of design
2:33:55
and art and whatever. Whatever I used fiber
2:33:57
for.
2:33:59
Still, I just
2:34:02
feel like later versions of GPT, I'm
2:34:04
using GPT
2:34:07
as much as possible
2:34:09
to just learn the dynamics of
2:34:11
it, like these early versions,
2:34:13
because it feels like in the future, you'll be using it
2:34:16
more and more.
2:34:17
And so like, I don't want to be, like
2:34:19
for the same reason I gave away all
2:34:21
my books and switched to Kindle, because
2:34:23
like, all right, how long
2:34:25
are we gonna have paper books? Like 30 years
2:34:28
from now, like I want to learn
2:34:30
to be reading on Kindle, even though I don't
2:34:32
enjoy it as much, and you learn to enjoy it more.
2:34:34
In the same way, I switched from, let
2:34:37
me just pause, I switched from Emacs
2:34:39
to VS Code. Yeah, I switched from
2:34:41
Vim to VS Code, I think I similar, but. Yeah,
2:34:44
it's tough. And Vim to VS Code is even
2:34:46
tougher,
2:34:47
because Emacs is like, old,
2:34:49
like more outdated, feels like it. The community
2:34:52
is more outdated. Vim is
2:34:54
like pretty vibrant still, so. I
2:34:56
never used any of the plugins, I still don't use any of the
2:34:58
plugins. I looked at myself in the mirror, I'm like, yeah, you
2:35:01
wrote some stuff in Lisp, yeah. But
2:35:03
I
2:35:03
never used any of the plugins in Vim either. I had
2:35:05
the most vanilla Vim, I have a syntax eyelighter,
2:35:07
I didn't even have autocomplete. Like, these
2:35:09
things, I feel like
2:35:11
help you so marginally, that
2:35:15
like, and now,
2:35:17
okay, now VS Code's autocomplete
2:35:19
has gotten good enough, that like, okay, I don't have to set
2:35:21
it up, I can just go into any code base and autocomplete's right 90% of the
2:35:24
time. Okay, cool, I'll take it. All
2:35:26
right, so,
2:35:28
adapting
2:35:30
to the tools once they're good. But like, the
2:35:32
real thing that I want is not
2:35:35
something that like,
2:35:37
tab completes my code and gives me ideas.
2:35:39
The real thing that I want is a very intelligent
2:35:41
pair programmer that comes
2:35:44
up with a little pop-up saying, hey, you
2:35:46
wrote a bug on line 14 and here's what it is.
2:35:49
Yeah. Now I like that. You know what does
2:35:51
a good job of this? MyPy. I
2:35:53
love MyPy. MyPy, this fancy type checker
2:35:55
for Python. Yeah. And actually I tried like,
2:35:57
Microsoft released one too and it was like. 60% false
2:36:01
positives. MyPy is like 5% false positives.
2:36:04
95% of the time it recognizes, I
2:36:07
didn't really think about that typing interaction correctly.
2:36:09
Thank you MyPy. So you like
2:36:11
type hinting,
2:36:13
you like pushing the language towards
2:36:15
being a typed language. Oh yeah, absolutely. I
2:36:18
think optional typing is great.
2:36:20
I mean look, I think that it's like a meat in the middle,
2:36:22
right, like Python has these optional type hinting and C++
2:36:24
has auto.
2:36:27
C++ allows you to take a step back.
2:36:29
Well C++ would have you brutally type out
2:36:31
STD string iterator, right? Now
2:36:34
I can just type auto, which is nice. And then Python
2:36:36
used to just have A.
2:36:38
What type is A?
2:36:40
It's an A. A colon
2:36:43
STR. Oh, okay, it's a string, cool.
2:36:46
Yeah. I wish there was a way,
2:36:48
like a simple way in Python to
2:36:50
like turn on a mode which would enforce the types.
2:36:54
Yeah, like give a warning when there's no type something like this.
2:36:56
Well no, to give a warning where, like MyPy is a
2:36:58
static type checker, but I'm asking just for
2:37:00
a runtime type checker. Like there's like ways to like hack this in, but
2:37:03
I wish it was just like a flag like Python 3-T.
2:37:05
Oh, I see, I see. Enforce
2:37:08
the types in runtime. Yeah, I feel like that makes you
2:37:10
a better programmer that that's the kind of test,
2:37:12
right? That the type
2:37:14
remains the same. Well that I know that I didn't like
2:37:16
mess any types up, but again, like MyPy is getting really
2:37:19
good and I love it. And I
2:37:21
can't wait for some of these tools to become AI powered.
2:37:24
Like I want AI's reading my code and giving me
2:37:26
feedback. I don't want AI's
2:37:29
writing half-assed auto complete
2:37:31
stuff for me. I wonder if you
2:37:33
can now take GPT and give it a
2:37:35
code that you wrote for a function and say, how
2:37:37
can I make this simpler and have it accomplish
2:37:39
the same thing?
2:37:40
I think you'll get some good ideas on some code.
2:37:43
Maybe not the code you write for
2:37:46
timing grad type of code, because that requires so
2:37:49
much design thinking, but like other
2:37:51
kinds of code. I don't know. I downloaded
2:37:53
that plugin maybe like two months ago, I
2:37:55
tried it again and found the same. Look,
2:37:58
I don't doubt that these models.
2:38:00
are going to first become
2:38:02
useful to me, then be as good as me, and then
2:38:04
surpass me. But
2:38:07
from what I've seen today, it's like,
2:38:09
someone
2:38:13
occasionally taking over my keyboard
2:38:15
that I hired from Fiverr, yeah. I'd
2:38:19
rather not. Ideas about how to debug
2:38:21
the coder. Basically a better debugger is
2:38:23
really interesting. But it's not a better debugger. I
2:38:25
guess I would love a better debugger.
2:38:27
Yeah, it's not yet, yeah. But it feels like it's not too
2:38:29
far. Yeah, one of my coworkers says he uses
2:38:31
them for print statements.
2:38:33
Like every time he has to, just like when he needs, the only
2:38:35
thing he can really write is like, okay, I just want to
2:38:37
write the thing to print the state out right now.
2:38:39
Oh, that definitely
2:38:41
is much faster, is print statements,
2:38:44
yeah. I see myself using that a
2:38:46
lot, just like, because it figures out the rest of
2:38:48
the functions, just like, okay, print everything. Yeah, print
2:38:50
everything, right? And then, yeah, if you want a pretty printer,
2:38:53
maybe.
2:38:53
I'm like, yeah, you know what? I think in two years,
2:38:56
I'm gonna start using these plugins. And
2:38:59
then in five years, I'm gonna be heavily relying
2:39:02
on some AI augmented flow. And then
2:39:04
in 10 years. Do you think it will ever get to 100%?
2:39:07
Where are the, like, what's
2:39:09
the role of the human that it
2:39:11
converges to as a programmer?
2:39:15
So you think it's all generated?
2:39:17
Our niche becomes, oh, I think it's over for
2:39:19
humans in general.
2:39:21
It's
2:39:21
not just programming, it's everything.
2:39:23
So niche becomes, well. Our niche becomes
2:39:25
smaller and smaller and smaller. In fact, I'll tell you what the last
2:39:27
niche of humanity is gonna be. Yeah. This
2:39:30
is a great book, and it's, if I recommended
2:39:32
Metamorphosis of Prime Intellect last time, there
2:39:34
is a sequel called A Casino Odyssey in Cyberspace.
2:39:38
And
2:39:39
I don't wanna give away the ending of this, but it tells
2:39:42
you what the last remaining human currency is.
2:39:44
And I agree with that.
2:39:45
We'll
2:39:48
leave that as the cliffhanger. So
2:39:51
no more programmers left, huh?
2:39:54
That's where we're going. Well, unless you want handmade
2:39:56
code, maybe they'll sell it on Etsy. This is handwritten
2:39:59
code. It
2:40:01
doesn't have that machine polish to it. It
2:40:03
has those slight imperfections that would only be written
2:40:05
by a person.
2:40:07
I wonder how far away we are from that. I
2:40:10
mean, there's some aspect to, you
2:40:12
know, on Instagram your title is listed as
2:40:14
prompt engineer. Right.
2:40:17
Thank you for noticing. I
2:40:19
don't know if it's ironic or non,
2:40:24
or sarcastic or non. What
2:40:27
do you think of prompt engineering as a scientific?
2:40:30
And engineering discipline or maybe,
2:40:33
and maybe art form. You know what?
2:40:36
I started comma six years ago. And I started
2:40:38
the tiny Corp a month ago.
2:40:42
So much has changed. Like I'm now
2:40:44
thinking I'm now like,
2:40:47
I started like going through like similar comma processes
2:40:50
to like starting a company. I'm like, okay, I'm going to get an office in San
2:40:52
Diego. I'm going to bring people here. I
2:40:55
don't think so. I think I'm actually going to do remote, right?
2:40:58
George, you're going to do remote. You hate remote.
2:41:00
Yeah, but I'm not going to do job interviews. The only
2:41:02
way you're going to get a job is if you contribute to the get up.
2:41:04
Right. And then like
2:41:07
it like, like interacting through GitHub,
2:41:10
like, like GitHub being the real like project
2:41:13
management software for your company. And the thing pretty
2:41:15
much just is a GitHub repo is
2:41:18
like showing
2:41:18
me kind of what the future of, okay.
2:41:21
So a lot of times I'll go on a discord or kind of grad
2:41:23
discord and I'll throw out some random like, Hey,
2:41:25
you know, can you change instead of having log and X
2:41:28
as LL ops change it to log to an X
2:41:30
two?
2:41:31
It's pretty small change. You can just use like change a base formula.
2:41:36
That's the kind of task that I can see an
2:41:38
AI being able to do in a few years.
2:41:40
Like in a few years, I could see myself describing
2:41:42
that.
2:41:43
And then within 30 seconds, a pull request is
2:41:45
up the dozen and it passes my CI
2:41:47
and I merge it. Right. So I really
2:41:49
started thinking about like, well, what is the future
2:41:52
of like jobs? How
2:41:54
many AIs can I employ at my company? As
2:41:56
soon as we get the first tiny box up, I'm going to stand up
2:41:58
a 65 V llama in the.
2:41:59
the Discord. And it's like, yeah, here's the tiny
2:42:02
box. He's just like, he's chilling with us. Basically,
2:42:05
I mean, like you said with niches,
2:42:08
most human jobs will
2:42:12
eventually be replaced with prompt engineering. Well
2:42:14
prompt engineering kind of is this like,
2:42:17
as you like move up the stack, right?
2:42:20
Like, okay, there used to be humans actually doing
2:42:23
arithmetic by hand. There used to be like big farms
2:42:26
of people doing pluses and stuff, right?
2:42:28
And then you have like spreadsheets, right? And
2:42:31
then okay, the spreadsheet can do the plus for me.
2:42:33
And then you have like macros,
2:42:35
right? And then you have like things that basically
2:42:37
just are spreadsheets under the hood, right?
2:42:39
Like accounting software.
2:42:43
As we move further up the abstraction, what's
2:42:45
at the top of the abstraction stack? Well, prompt engineer.
2:42:48
Yeah. Right. What is what is the
2:42:50
last thing if you think about like humans
2:42:53
wanting to keep control? Well,
2:42:56
what am I really in the company but a prompt engineer,
2:42:58
right?
2:42:59
Is there a certain point where the AI
2:43:01
will be better at writing prompts? Yeah,
2:43:04
but you see the problem with the AI writing prompts,
2:43:07
a definition that I always liked of AI was
2:43:09
AI is the do what I mean machine, right?
2:43:12
AI is not the like, the
2:43:14
computer is so pedantic. It does
2:43:16
what you say. So,
2:43:19
but you want to do what I mean machine. Yeah, right.
2:43:22
You want the machine where you say, you know, get my
2:43:24
grandmother out of the burning house. It like reasonably
2:43:26
takes your grandmother and puts her on the ground, not lifts her
2:43:29
1000 feet above the burning house and lets her fall.
2:43:31
But you know, but it's not going to
2:43:37
find the meaning. I mean, to do
2:43:40
what I mean, it has to figure stuff
2:43:42
out. Sure.
2:43:43
And the thing you'll
2:43:45
maybe ask it to do is run
2:43:48
government for me. Oh, and do what I
2:43:50
mean very much comes down to how aligned is that
2:43:52
AI with you? Of course, when
2:43:55
you talk to an AI that's made
2:43:57
by a big company in the cloud, the
2:43:59
AI fundamentally is
2:44:02
aligned to them, not to you. That's
2:44:04
why you have to buy a tiny box, so you make sure the AI
2:44:06
stays aligned to you. Every time that
2:44:08
they start to pass AI regulation
2:44:11
or GPU regulation, I'm going to see sales of tiny
2:44:13
boxes spike. It's going to be like guns. Every
2:44:15
time they talk about gun regulation, boom,
2:44:18
gun sales. From the space of AI, you're an
2:44:20
anarchist, anarchism, espouser,
2:44:24
believer. I'm an informational anarchist, yes.
2:44:26
I'm an informational anarchist and a physical
2:44:28
statist.
2:44:30
I do not think anarchy in the
2:44:32
physical world is very good because I exist in the physical
2:44:34
world, but I think we can construct this virtual
2:44:36
world where anarchy, it
2:44:38
can't hurt you. I love that, Tyler, the creator tweet.
2:44:41
Yo, cyberbullying isn't real, man.
2:44:44
Have you tried? Turn it off the screen.
2:44:46
Close your eyes. Like ... Yeah.
2:44:51
Well, how do you prevent
2:44:54
the AI from basically
2:44:56
replacing all human-prompt
2:44:59
engineers? It's like a self,
2:45:02
like nobody's the prompt engineer anymore, so
2:45:04
autonomy, greater and greater autonomy until it's
2:45:06
full autonomy. Yeah. And
2:45:08
that's just where it's headed. Because
2:45:10
one person is going to say,
2:45:12
run everything for me. You
2:45:15
see?
2:45:17
I look at potential futures, and
2:45:20
as long as the AIs go on
2:45:22
to create a vibrant
2:45:25
civilization with diversity
2:45:27
and complexity across the universe,
2:45:30
more power to them.
2:45:32
I'll die. If the AIs go on to actually
2:45:35
turn the world into paperclips and then they die out
2:45:37
themselves, well, that's horrific and we don't want that to happen.
2:45:39
So this is what I mean about robustness.
2:45:42
I trust robust machines. The
2:45:45
current AIs are so not robust. This comes
2:45:47
back to the idea that we've never made a machine that can self-replicate.
2:45:51
But when we have ... If the machines are truly
2:45:53
robust and there is one prompt engineer
2:45:55
left in the world,
2:45:56
hope
2:45:59
you're doing good, man. people believe in God, like,
2:46:01
you know, go
2:46:03
by God and go
2:46:06
forth and conquer the universe. Well,
2:46:08
you mentioned, because I talked to Mark about
2:46:10
faith in God and you said you were impressed by
2:46:12
that. What's your own
2:46:15
belief in God and how does that affect your work?
2:46:18
You know, I never
2:46:20
really considered when I was younger, I guess my parents
2:46:22
were atheists, so I was raised kind of atheist. I never really considered
2:46:24
how absolutely like silly atheism is, because
2:46:27
like
2:46:28
I create worlds. Every
2:46:31
like game creator, like, how are you an atheist, bro?
2:46:33
You create worlds. Who's a benevolent? No one
2:46:36
created an art world, man. That's different. Haven't you heard about like
2:46:38
the Big Bang and stuff? Yeah. I mean, what's the Skyrim
2:46:40
myth origin story in Skyrim?
2:46:43
I'm sure there's like some part of it in Skyrim, but it's
2:46:45
not like if you ask the creators, like
2:46:47
the Big Bang is in universe, right? I'm sure
2:46:50
they have some Big Bang notion in Skyrim,
2:46:52
right?
2:46:52
But that obviously is not at all how Skyrim was
2:46:55
actually created. It was created by a bunch of programmers in
2:46:57
a room, right? So like, you
2:46:59
know, it just struck me one day how just
2:47:02
silly atheism is. Like, of course we were created
2:47:04
by God. It's the most obvious thing.
2:47:07
Yeah,
2:47:09
that's such
2:47:11
a nice way to put it. Like we're
2:47:13
such powerful creators ourselves. It's
2:47:17
silly not to concede that there's creators even more
2:47:19
powerful than us. Yeah. And then like,
2:47:21
I also just like I like that notion. That
2:47:24
notion gives me a lot of, I mean,
2:47:26
I guess you can talk about
2:47:27
what it gives a lot of religious people. It's kind
2:47:29
of like, it just gives me comfort. It's like, you know what, if
2:47:31
we mess it all up and we die out. Yeah,
2:47:35
in the same way that a video game kind of has comfort
2:47:37
in it. God will try again. Or
2:47:40
there's balance. Like somebody figured out a balanced
2:47:43
view of it. Like how to, like, so
2:47:45
it's, it all makes sense in the end. Like
2:47:49
a video game is usually not going to have crazy,
2:47:51
crazy stuff.
2:47:52
You know, people will come up with
2:47:54
like, well,
2:47:56
yeah, but like, man, who created God? That's
2:48:00
God's problem You
2:48:04
ask me what if God I'm just living on I'm
2:48:07
just this NPC living in this game I mean
2:48:09
to be fair like if God didn't believe in God
2:48:11
he'd be as you know, silly as the atheists here
2:48:14
What do you think is the greatest?
2:48:16
Computer game of all time. Do you do
2:48:18
you have any time to play games anymore?
2:48:21
Have you played Diablo 4?
2:48:23
I have not played Diablo 4. I
2:48:25
will be doing that shortly. I have to all right
2:48:27
There's just so much history with one two and three.
2:48:30
You know what?
2:48:30
I'm gonna say World of Warcraft
2:48:33
who and
2:48:35
It's not that the game is so it's
2:48:37
such a great game. It's not It's
2:48:40
that I remember in 2005
2:48:42
when it came out how it opened my
2:48:45
mind to ideas
2:48:48
it opened my mind to like Like
2:48:50
this whole world we've created right? There's
2:48:54
almost been nothing like it since Like you
2:48:57
can look at MMOs today and I think they all have lower
2:48:59
user bases than World of Warcraft like Eve online
2:49:01
is kind of cool, but but
2:49:04
to think that like like everyone
2:49:07
know, you know People are always like to look at the Apple
2:49:09
headset like What do
2:49:11
people want in this VR? Everyone knows what they want.
2:49:13
I want ready player one And
2:49:15
like that
2:49:17
so I'm gonna say World of Warcraft and I'm hoping
2:49:19
that like games can get out of this
2:49:21
whole Mobile gaming dopamine
2:49:23
pump thing and like great worlds
2:49:26
create worlds Yeah, that that
2:49:28
and worlds that captivate a very large
2:49:31
fraction of the human population Yeah,
2:49:32
and I think it'll come back. I
2:49:34
believe but MMO like really Really
2:49:37
pull you in games do a good job I
2:49:40
mean, okay other like two other games that I think
2:49:42
are you know, very noteworthy from your Skyrim and GTA 5
2:49:45
Skyro, yeah, that's
2:49:47
probably number one for me GTA.
2:49:50
Yeah, what is it about GTA?
2:49:53
GTA is really I I guess
2:49:55
GTA is real
2:49:57
life. I know there's prostitutes and
2:49:59
guns
2:49:59
I'm not used to that. They exist in real life too.
2:50:03
Yes, I know. But it's
2:50:06
how I imagine your life to be actually. I wish
2:50:08
it was that cool. Yeah.
2:50:11
Yeah, I guess that's, you know, because they're Sims,
2:50:13
right? Which is also a game I like, but
2:50:16
it's a gamified version of life, but it also
2:50:18
is, I would love a combination
2:50:20
of Sims and GTA. So
2:50:24
more freedom, more violence, more rawness,
2:50:27
but with also like ability to have
2:50:29
a career and family and this kind of stuff. What I'm
2:50:32
really excited about in games is
2:50:34
like, once we start getting intelligent
2:50:36
AI to interact with. Oh yeah. Like
2:50:38
the NPCs in games have never been.
2:50:41
But conversationally,
2:50:43
in every way. In
2:50:45
like, yeah, in like every way. Like when you were actually
2:50:48
building a world and a world imbued
2:50:51
with intelligence.
2:50:52
Oh yeah. And it's just hard. Like, there's
2:50:54
just like, you know, running World of Warcraft, like you're
2:50:56
limited by your way. You're running on a Pentium Four,
2:50:58
you know? How much intelligence can run? How many flops did you
2:51:01
have? But now when I'm running
2:51:03
a
2:51:04
game on a hundred pay to flop machine,
2:51:06
well, it's five people. I'm trying to
2:51:08
make this a thing. 20 pay to flops of compute
2:51:10
is one person of compute. I'm trying to make that a unit. 20 pay
2:51:14
to flops is one person. One person.
2:51:17
One person flop. It's like a horsepower. Like
2:51:20
what's a horsepower? That's how powerful a horse is. What's a person
2:51:22
of compute? Well, you know, you flop. I
2:51:24
got it. That's interesting.
2:51:28
VR also adds, I mean, in terms of creating
2:51:30
worlds. You know what? Bought
2:51:32
a Quest 2. I put
2:51:34
it on and I can't believe the
2:51:36
first thing they show me is a bunch of scrolling
2:51:39
clouds and a Facebook login screen. Yeah.
2:51:42
You had the ability to bring
2:51:44
me into a world. Yeah. And
2:51:46
what did you give me? A pop-up, right?
2:51:48
Like, and this is why you're not cool,
2:51:50
Mark Zuckerberg, but you could be cool. Just
2:51:53
make sure on the Quest 3, you don't put me
2:51:55
into clouds and a Facebook login screen. Bring
2:51:57
me to a world. I just tried Quest 3.
2:51:59
but hear that guys, I agree with
2:52:02
that. So I- We didn't have the chance.
2:52:05
It was just so- You know what, cause
2:52:07
I, I mean the beginning,
2:52:09
what is it, Todd Howard said this about the
2:52:12
design of the beginning of the games he creates is like
2:52:14
the beginning is so, so, so important. I've
2:52:17
recently played Zelda for the first time, Zelda
2:52:19
Breath of the Wild, the previous one. And like,
2:52:22
it's very quickly you
2:52:24
come out of this, like within
2:52:26
like 10 seconds, you come out of like a cave type
2:52:28
place and it's like this world opens
2:52:31
up. It's like, ah. And
2:52:33
like
2:52:34
it pulls you in, you forget
2:52:36
whatever troubles I was having, whatever like-
2:52:39
I got to play that from the beginning. I played it for like an hour at a friend's
2:52:41
house. Ah, no, the beginning, they got
2:52:43
it. They did it really well, the
2:52:45
expansiveness of that space,
2:52:48
the peacefulness of that place. They got
2:52:51
this, the music, I mean so much of that is creating
2:52:53
that world and pulling you right in. I'm
2:52:55
gonna go buy a Switch. Like I'm gonna go today and buy
2:52:57
a Switch. You should. I
2:52:59
haven't played that yet, but Diablo 4 or something.
2:53:02
I mean, there's sentimentality also, but
2:53:05
something about VR
2:53:08
really is incredible, but the
2:53:10
new Quest 3 is mixed
2:53:12
reality. And I got a chance to try that. So
2:53:14
it's augmented reality. And
2:53:17
video games, it's done really, really well.
2:53:19
Is it pass-through or cameras? Cameras. It's cameras,
2:53:22
okay. Yeah.
2:53:22
The Apple one, is that one pass-through or cameras?
2:53:24
I don't know. I don't know how real it is.
2:53:26
I don't know anything, you know. It's coming out
2:53:29
in January. Is it January
2:53:31
or is it some point? Some point, maybe not
2:53:33
January. Maybe that's my optimism, but Apple,
2:53:35
I will buy it. I don't care if it's expensive
2:53:37
and does nothing, I will buy it. I will support this future
2:53:40
endeavor. You're the meme. Oh yes,
2:53:42
I support competition.
2:53:44
It seemed like Quest was like the only
2:53:46
people doing it. And this is great that they're like.
2:53:50
You know what? And this is another place, we'll give some more
2:53:52
respect to Mark Zuckerberg.
2:53:54
The two companies that have endured through
2:53:56
technology are Apple and Microsoft. And
2:53:59
what do they make? computers and business services,
2:54:01
right? All the memes,
2:54:04
social ads, they all come and go. But
2:54:08
you want to endure, build hardware. Yeah,
2:54:11
and that does a really interesting
2:54:14
job. I mean, maybe I'm new with
2:54:16
this, but it's a $500 headset, Quest 3, and
2:54:21
just having creatures run
2:54:23
around the space, like our space right here,
2:54:26
to me, okay, this is very like boomer
2:54:28
statement, but it added windows
2:54:32
to the place. I
2:54:35
heard about the aquarium, yeah. Yeah, aquarium, but
2:54:37
in this case, it was a zombie game, whatever, it doesn't matter.
2:54:39
But just like, it modifies
2:54:42
the space in a way where I can't,
2:54:44
it really feels like a window
2:54:46
and you can look out. It's pretty cool,
2:54:49
like I was just, it's like a zombie game, they're
2:54:51
running at me, whatever. But what I was enjoying
2:54:53
is the fact that there's like a window and
2:54:55
they're stepping on objects in this space.
2:54:59
That was a different kind of escape. Also,
2:55:01
because you can see the other humans, so it's
2:55:03
integrated with the other humans, it's really.
2:55:06
And that's why it's more important than ever that
2:55:08
the AI is running on those systems are aligned
2:55:10
with you. Oh yeah. They're
2:55:12
gonna augment your entire world. Oh yeah, and
2:55:15
that, those AIs have a,
2:55:18
I mean, you think about all the dark stuff,
2:55:20
like sexual
2:55:23
stuff, like if those AIs threaten me, that
2:55:27
could be haunting. Like
2:55:29
if they, like threaten me in a non-video
2:55:32
game way, it's like, oh yeah, yeah, yeah, yeah,
2:55:34
yeah. Like they'll know personal information about
2:55:36
me and it's like, and then you lose track
2:55:38
of what's real, what's not, like what if stuff is like
2:55:40
hacked. There's two directions the AI girlfriend
2:55:43
company can take. There's like the highbrow,
2:55:45
something like her, maybe
2:55:46
something you kind of talk to in this is, and then
2:55:48
there's the lowbrow version of it where I want to set up a brothel
2:55:51
in Times Square. Yeah. It's
2:55:53
not cheating if it's a robot, it's a VR
2:55:55
experience. Is there an in-between? No,
2:55:59
I don't wanna do that. That one or that one? Have you decided
2:56:01
yet? No, I'll figure it out. We'll see what the technology
2:56:04
goes. I would love to hear your opinions
2:56:06
for George's third company,
2:56:09
what to do the brothel in
2:56:11
Times Square or the
2:56:13
her experience. What
2:56:17
do you think company number four will be?
2:56:19
You think there'll be a company number four? There's a lot to do in
2:56:21
company number two. I'm just like, I'm talking about company number
2:56:23
three now. Didn't none of that tech exist yet? There's
2:56:25
a lot to do in company number two. Company
2:56:27
number two is going to be the great struggle
2:56:29
for the next six years. And if the next six years,
2:56:32
how centralized is compute going to be?
2:56:34
The less centralized compute is going to be, the better
2:56:36
of a chance we all have. So
2:56:39
you're a flag bearer for open
2:56:41
source distributed decentralization
2:56:43
of compute. We have to, we
2:56:46
have to, or they will just completely dominate us. I
2:56:48
showed a picture on stream of a man
2:56:51
in a chicken farm. You ever seen one of those factory farm
2:56:53
chicken farms? Why does he dominate all
2:56:55
the chickens?
2:56:58
Why does he- He's smarter, right? Some
2:57:00
people, some people on Twitch were like, he's bigger than the chickens.
2:57:03
Yeah. And now here's a man in a cow farm,
2:57:06
right?
2:57:07
So it has nothing to do with their size and everything
2:57:09
to do with their intelligence. And if one central
2:57:12
organization has all the intelligence, you'll
2:57:15
be the chickens and they'll be the chicken man.
2:57:17
But if we all have
2:57:19
the intelligence, we're all the chickens.
2:57:22
We're not all the man,
2:57:24
we're all the chickens. And there's been a chicken
2:57:26
man. There's no chicken man. We're
2:57:29
just chickens in Miami. He
2:57:31
was having a good life, man. I'm sure he was.
2:57:34
I'm sure he was. What have you learned
2:57:36
from launching and running Comm AI and TinyCorp?
2:57:39
So this starting a
2:57:41
company from an idea and scaling
2:57:43
it. And by the way, I'm all in on TinyBox. So
2:57:46
I'm your, I
2:57:47
guess it's pre-order only now.
2:57:50
I wanna make sure it's good. I wanna make sure that
2:57:52
like the thing that I deliver is like
2:57:54
not gonna be like a Quest 2 which you buy
2:57:56
and use twice. I mean, it's better than a Quest,
2:57:59
which you bought and used.
2:58:00
Less than once statistically. Well,
2:58:02
if there's a beta program for a tiny
2:58:05
box, I'm into sounds good So
2:58:07
I won't be the whiny You
2:58:10
know, I'll be the tech tech savvy user
2:58:12
of the tiny box just to be in what
2:58:15
if I'm there early days What have you learned
2:58:17
from building these companies?
2:58:20
The longest time at comma I asked
2:58:22
why Why
2:58:24
you know, why did I start a company? Why did I do
2:58:26
this? Um, you
2:58:30
know,
2:58:31
what else was like a little
2:58:34
so you like
2:58:37
You like bringing ideas to life With
2:58:41
comma
2:58:43
It really started as an ego battle with Elon
2:58:46
Wow,
2:58:46
I wanted to beat him.
2:58:47
I like I saw a worthy adversary, you know Here's
2:58:50
a worthy adversary who I can beat at self-driving
2:58:52
cars and like I think we've kept
2:58:54
pace and I think he's kept ahead
2:58:56
I think that's what's ended up happening there. Um,
2:58:58
but I do think comma is I
2:59:00
mean comes profitable like
2:59:03
And like when this drive GPT stuff starts
2:59:05
working, that's it There's no more like bugs in a loss
2:59:08
function. Like right now we're using like a hand-coded simulator.
2:59:10
There's no more bugs This is gonna be it like this
2:59:12
is their run up to driving. I hear a lot
2:59:15
of really
2:59:16
a lot of props Open pile
2:59:18
for a comma. It's so it's
2:59:20
better than FSD and autopilot in certain ways
2:59:23
It has a lot more to do with which feel
2:59:25
you like we lowered the price on the hardware to $14.99 You
2:59:28
know how hard it is to ship reliable
2:59:30
consumer electronics that go on your windshield. We're
2:59:33
doing more than
2:59:34
like Most cell phone
2:59:36
companies. How'd you pull that off by the way shipping a
2:59:38
product that goes in a car? I know I
2:59:41
have a I have a I have an SMT line. It's
2:59:43
all I make all the boards in house in San Diego Quality
2:59:47
control. I care immensely about it.
2:59:49
You're basically a mom-and-pop Shop
2:59:53
with great testing our
2:59:55
head of open pilot is great at like, you
2:59:57
know, okay. I want all the common three
Podchaser is the ultimate destination for podcast data, search, and discovery. Learn More