Episode Transcript
Transcripts are displayed as originally observed. Some content, including advertisements may have changed.
Use Ctrl + F to search
0:01
Today on the AI Daily Brief, everything
0:03
announced at Google I.O. and a look at
0:05
the battle between Google and OpenAI.
0:08
The AI Daily Brief is a daily podcast and
0:10
video about the most important news and discussions in
0:12
AI. To join the conversation, hop
0:14
on the Discord with the link in our show notes. Hello
0:25
friends, one more day of non-traditional formats we're
0:27
just going to be talking about Google I.O.
0:29
and the comparison to OpenAI as that's where
0:31
a ton of conversation is right now. The
0:33
other big obvious story that we will cover
0:36
is Ilya formally leaving OpenAI, so keep your
0:38
ears peeled for that in the next day
0:40
or so. But for now, let's talk about
0:42
Google I.O. Welcome
0:44
back to the AI Daily Brief. Yesterday was
0:46
the main keynote day of Google I.O. This is
0:48
Google's big annual conference where they announce all the
0:51
stuff that they're working on and at the surprise
0:53
of no one AI was even more center stage
0:55
than it was last year. Now the
0:57
timing of this event was no coincidence, or
0:59
more specifically the timing of OpenAI's event just
1:02
before this was no coincidence. It was pretty
1:04
clear then and is definitely clear now that
1:06
OpenAI was trying to front run Google and
1:08
capture excitement and attention before I.O. happened. What
1:11
we're going to do today is talk about
1:13
the difference between these two announcements, who the
1:15
AI Twitter sphere thinks won, but we have
1:17
to start with just everything that Google actually
1:19
announced. This is in no particular
1:21
order other than broadly speaking the biggest most
1:23
discussed things being up top. First
1:26
of all, we got a new version of Gemini. This is called Gemini
1:28
1.5 Flash. The
1:30
team at Google said that this multimodal model was as
1:32
powerful as Gemini 1.5 Pro, but that
1:34
it had been optimized for quote narrow
1:36
high frequency low latency tasks, making it
1:39
better at generating fast responses. They
1:41
also announced that the context window was moving from
1:43
the already industry leading one million up to two
1:45
million in the coming months. Second
1:47
and perhaps most relevant for the demos that
1:50
we saw a couple days ago from OpenAI,
1:52
we got Project Astra. This is their version
1:54
of a personal assistant. On screen
1:56
is the demo video they shared both live and on
1:58
socials. What?
2:01
In that part of the speaker called.
2:05
That. Is the tweeter. It produces
2:07
high frequency sounds. The. Verge
2:09
Rights project. Astra is powering many of the most
2:11
impressive demos for my oh this year, and the
2:13
company's aim for it is to be an honest
2:15
to goodness Ai agent that can't just talk to
2:17
you, but also actually does things on your behalf.
2:19
Will. Come back to that in a moment. given
2:22
that is such a strong and clear through
2:24
line across both the open A I presentation
2:26
and does Google presentation. These companies are betting
2:28
on a totally new mode of interaction between
2:30
people in A that they're literally betting their
2:32
whole enterprise on. We. Also got
2:34
Googles answer to Soar a. Couple.
2:36
Months ago there had been a question of whether
2:38
open A. I was starting to lose the lead
2:41
that it has held for so long, but then
2:43
they dropped a set of demos for Saura, and
2:45
honestly everyone was just slack jawed. It was so
2:47
far ahead of what was available currently from other
2:49
text video generators that it honestly shifted many people's
2:52
opinions on whether specialized models can ever be to
2:54
be generalist model. Then. Again, uncommonly for
2:56
Open A, I saw a hasn't been available
2:58
for people to test. Part of that seems
3:00
to be economics with generations from storage is
3:02
costing an absolute boatload to create right now.
3:05
Instead, Open Ai has been pursuing relationships with
3:07
Hollywood in professional filmmakers but with bill Google
3:09
has produced an answer like up in a
3:11
Eyesore are Be A was not available to
3:13
the public and instead Google has been working
3:15
with a set of high profile you to
3:18
creators as well as Hollywood filmmakers to try
3:20
to start integrating the technology. I'll
3:22
also saw the announcement of a competitor
3:24
to Open a Eyes customs you be
3:26
tease something that Google is calling gems.
3:28
They. Are basically putting it in exactly the same way
3:31
that those custom she keys were created. Their announcement
3:33
tweet reads: Whether you need a yoga best the
3:35
or a calculus tutor in the coming months, you'll
3:37
be able to customize Gemini saving time when you
3:39
have specific ways you interact with Gemini again and
3:41
again. Gemini. Like Cbd for got
3:43
a conversational upgrade for voice adding personality,
3:45
the ability to interrupt it, and lower
3:48
latency to make conversations hopefully feel more
3:50
natural. And then from there there were
3:52
a ton of announcements of weird Gemini
3:54
was being integrated far more deeply into
3:56
Google products that it happened before. For.
3:58
example gemini is getting much more deeply
4:01
integrated into workspace. The Verge
4:03
again writes, Google is rolling Gemini 1.5 Pro into the
4:05
sidebar for Docs, Sheets, Slides, Drive, and Gmail. When it
4:07
rolls out to paid subscribers next month, it will turn
4:09
into more of a general purpose assistant within workspace that
4:11
can fetch info from any and all of the content
4:14
from your drive, no matter where you are. It will
4:16
also be able to do things for you like write
4:18
emails and incorporate info from a document you're currently looking
4:20
at, or remind you later to respond to an email
4:23
you're perusing. Interestingly, this integration is starting
4:25
to get so deep that it sort of appears
4:27
to me that Google is making a bet on
4:29
a future where our default interaction with computers is
4:31
talking to our AI assistant that then goes and
4:34
does the stuff that we do now for us.
4:36
So instead of writing an email, you talk to
4:38
your assistant, which writes the email. Instead of analyzing
4:40
a document, you ask your AI assistant to analyze
4:42
a document, etc., etc. This
4:45
one is coming pretty soon, so we'll see
4:47
just how ubiquitous it actually becomes and whether
4:49
people really start to shift those behaviors. Another
4:51
big important integration is around search.
4:54
Thank goodness we're losing the language
4:56
of search-generative experiences, which is an
4:58
incredible mouthful, and instead they're being
5:00
replaced with what they're calling AI
5:02
overviews. Basically, every search now is
5:04
going to have a perplexity-style summarization
5:06
first, rather than just the classic
5:08
blue links. It is not hyperbolic
5:10
to call this the most significant shift to Google
5:13
search since the product was launched 20-plus years ago.
5:16
Relatedly, Google Chrome is getting an AI assistant,
5:18
which will basically place text generation natively in
5:20
the browser. As
5:23
a listener of this show, I have
5:25
a strong feeling you like to stay
5:27
up-to-date on all things artificial intelligence, including
5:29
its impact on the workforce, which is
5:31
why I highly recommend checking out Managing
5:33
the Future of Work, the chart-topping business
5:36
podcast from Harvard Business School. HBS professors
5:38
Bill Kerr and Joe Fuller talk to
5:40
business leaders, technologists, and policymakers grappling with
5:42
the forces like AI, globalization, and demographic
5:44
shifts that are reshaping the nature of
5:46
work. Recent guests include IBM CHRO, Nicole
5:49
Lamoureux on how Big Blue is adopting
5:51
AI, Morningstar CEO Kunal Kapoor
5:53
on how AI can raise the investment IQ,
5:56
Microsoft corporate vice president Jared Spitero on how
5:58
the tech giant is experimenting. its way
6:00
from AI assistance to autonomous agents and
6:02
many other prominent movers in business and
6:04
the workforce ecosystem. So don't miss out.
6:07
Follow Managing the Future of Work on Apple
6:09
Podcasts, Spotify, or wherever you're listening now. Hello
6:13
friends, quick note before we get back to
6:15
the show, I'm so excited to share that
6:18
Super Intelligent is now live. Super
6:20
Intelligent is a platform for fast, fun,
6:22
and super practical, useful AI learning. We
6:25
have something like 300 video
6:27
tutorials adding 30 to 50 each week,
6:29
covering every topic in AI you can
6:31
imagine from LLMs to image generators to
6:33
case studies, use cases, basically everything that
6:35
tells you how to use AI and
6:37
what to use it on. In short,
6:39
fast four to seven minute tutorial videos,
6:41
which are paired with step by step
6:44
instructions that help you actually use these
6:46
tools as well. It's $20
6:48
a month for unlimited access and I would love
6:50
to see you there. Check it out
6:52
at Bsuper.ai. So
6:57
let's talk about impressions. First
7:00
of all, I think Jan Pelec nailed what a lot of people were
7:02
feeling when he wrote, Google is out
7:04
for blood. What a bombardment. In
7:07
many ways, that's what it felt like to
7:09
me too, as opposed to open AI's tight
7:11
30 minute announcement. Google's IO keynote went on
7:13
for a couple of hours. In fact, in
7:15
its hugeness, it sort of gave the impression
7:17
that Google wasn't trying to compete one to
7:20
one between Project Astra, their AI assistant and
7:22
open AI's GPT-4.0 assistant. But instead to show
7:24
just how many different vectors of AI Google
7:26
is competing on. Still, there was
7:28
a lot of focus on comparing Project Astra
7:30
and GPT-4.0. Professor Ethan
7:33
Molloch writes, this video does a great
7:35
job highlighting the practical side of ubiquitous
7:37
multimodal AI agents. It's fascinating to
7:39
see the neck and neck technology race between Google and
7:41
open AI around the same set of problems and approaches.
7:44
Robert Lukosko writes, guys, that's really insane. I
7:46
rewatched this IO agents demo three times already.
7:48
Tell me Google will not win agent space.
7:51
It can schedule UPS pickup from a single
7:53
photo. However. The biggest critique was
7:55
that we're still comparing what was a slightly
7:57
janky live demo in the case of Open
7:59
AI. From extremely high production
8:01
demo videos prepared for google. A
8:04
mere from Do As Rights. there's a big
8:06
difference between Open A I Google of a
8:08
I ships useful stuff. For example, the Gb
8:10
for a model is incredibly fast and fifty
8:12
percent cheaper, and the voice improvement is incredible.
8:14
It's widely available, Google announced the stuff that
8:16
looks great and I owe presentations but isn't
8:18
really usable or even released. In many cases,
8:21
it's mostly vaporware. It's fantastic to see a
8:23
company with real potential to disrupt Google. B
8:25
C Investor rights. Google simply lacks credibility for
8:27
any kind of Ai demo at this point
8:29
like Project Ass or similar promo reels. Even
8:31
worse, meanwhile, Open A I does a semi
8:33
scripted event for Gb for oh, but enough
8:36
flaws come through to show it's real Google.
8:38
Need a full culture reset at this point?
8:40
What? Is referring to In part is that
8:42
people got extremely hyped back last December
8:44
when Google announced Gemini Ultra only to
8:46
realize that the videos that had been
8:49
most impressive. Really pretty doctor to show
8:51
exactly what they wanted to show vs
8:53
where the actual capacity was. Some.
8:55
People also pointed out that even these overly prepared
8:57
demos weren't as impressive as what Open A I
8:59
showed. Be ready! Who was a little bit harsh
9:01
on up at A I frankly wrote after watching
9:03
Google I would say to say what Open A
9:06
I showed yesterday was mind blowing. Astra is a
9:08
prototype voice assistant and seem like a two year
9:10
old baby compared to Away I, Scarlett Johanson. Daily.
9:13
Rodriguez said, well that's quite a turnaround from
9:15
yesterday to it's been to. Reply to definitely
9:17
is low Google I'll give me a comparison
9:19
and comparatively be open. A Iverson was very,
9:21
very good. Unfortunately, the people who
9:23
were at I oh and got their hands
9:25
on project Astra also weren't all that complementary.
9:28
Santiago. On twitter said I tried
9:30
Project Astra and the demo wasn't great.
9:33
One. Of the things that was totally on ignore
9:35
a bull was the difference in approach. Googles.
9:37
I'll presentation was, like I said, extremely
9:39
high production value and it also had
9:41
celebrity integrations woven throughout of an Ai.
9:43
Presentation was intimate and quaint by comparison. However,
9:45
I'm not so sure that does exactly
9:47
what Google one it. I asked my
9:49
Twitter followers to Googles high production values
9:51
and celebrity integrations make you more or less
9:54
stoked on their Ai announcements. literally. Forty
9:56
two point six percent of people said
9:58
it didn't impact their opinion, but. Those
10:00
who did have it affect their opinion about
10:02
ninety percent or less stoke rather than more
10:04
stoked. Broadly. Speaking other polls I
10:06
saw confirmed a pretty similar sensibility about
10:08
the two events. Daedalus asked did Google
10:10
cook open a I seventy five point
10:12
Nine percent of respondents said not Stanford
10:14
Enter Gao wrote who wins they Today
10:16
among nearly sixteen hundred voters. Sixty point
10:18
four percent said open A I'd compared
10:21
to just sixteen point One percent who
10:23
said Google. Jedi. I who does
10:25
research at Microsoft wrote so who on the
10:27
week and of her four hundred respondents, eighty
10:29
seven point five percent said open A I.
10:31
It. Is worth noting, however, that there
10:33
may be an audience bias issue here.
10:36
Open. Ai is very much competing for the
10:38
state of the art they are, as they
10:41
have been leading the pack. From a technology
10:43
perspective, a significant portion of the people who
10:45
are discussing this on a I, Twitter or
10:47
themselves technologist, developers and entrepreneurs. They're the type
10:50
of people in other words who value State
10:52
of the Art more than anything else. That's.
10:54
Different. I think that the audience that
10:56
Google is going for in the strategy
10:58
that Google is taking Google is taking
11:01
an approach of putting a I everywhere
11:03
with their incredibly large and diverse install
11:05
base. Open. A I does not
11:07
have users that they are bringing to
11:09
their party. Google has everyone using G,
11:11
email, everyone using maps, everyone using search,
11:14
everyone using docs, everyone using slides, etc,
11:16
etc, etc. While. They are clearly not
11:18
interested in seeding stated the art in any
11:20
way. the strategy that they're actually deploying doesn't
11:22
technically rely on being stated the art it
11:25
requires being good enough for these products to
11:27
actually be useful for the people who are
11:29
already there users. It's frankly a much lower
11:31
burden and one of the reasons that even
11:33
for an open A I, it's so incredibly
11:35
hard to compete in the space. So.
11:37
To the extent that you are trying to find
11:39
signal in these results, I don't think it's about
11:41
who is necessarily leading a I in the real
11:43
world, but it does seem to be clear that
11:45
people think that what open a eyes producing is
11:47
still a little bit at least ahead of what
11:49
Google Labs are producing. For. Some now they're
11:52
not ready to sign off on any of these yet.
11:54
Robert. scoble right time and a judge he moves of here
11:56
some things i'm going to do with my ai personal assistant since
11:58
that is what open a i and Google are trying to bring
12:00
to us. Scoble then puts together a list
12:02
of 30 questions by which he will
12:05
judge an AI assistant. For example, he writes, does it
12:07
do the basics? Like when you ask Siri to take
12:09
you back to your car, it doesn't work. But if
12:11
you ask it to take you back to your parked
12:13
car, it does. If you ask it
12:15
a thousand questions, how many does it answer accurately? In
12:17
my perfect world, it would have no mistakes, etc, etc,
12:20
etc. And I think,
12:22
hold aside Scoble's specifics, what he's getting at is that
12:24
ultimately AI assistants will only be a thing if they're
12:26
good enough for them to be a thing. The
12:28
general mass market of consumers isn't going to use
12:30
something because it's AI and cool. They're going to
12:32
use something strictly on the basis of whether it's
12:34
useful. We're still at the very early
12:37
stages of these products coming to market and
12:39
it's not at all clear to me yet
12:41
that either Project Astra or the new GPT-40
12:43
is actually going to clear that threshold. Still
12:46
after all of this, the competitive landscape is getting
12:48
a little bit clearer. DC Investor again pointed out
12:50
open AI stakes and why the news about them
12:52
potentially getting a deal with app is even more
12:54
interesting today than it was a couple of days
12:56
ago. He writes, the big thing
12:58
I see with Google's AI announcement yesterday is they
13:00
are driving for much deeper integration with your data
13:03
and your surroundings. Open AI better hope Apple can
13:05
do the same for them. I could see an
13:07
argument that open AI needs Apple as much as
13:09
Apple needs open AI. At the
13:12
same time, there are clearly some forces that have
13:14
been unleashed by AI that all
13:16
of these companies now feel subject to.
13:18
Darrell Bossenjo writes, the real problem open AI
13:20
has created for Google is that it has damaged
13:22
the virtuous cycle of websites making free content for
13:25
Google to index so Google sends them traffic in
13:27
return. Now Google AI results will provide
13:29
answers instead of sending sites traffic and the web
13:31
withers. Pete Pashall put it more
13:33
simply, it's official, generative search is now the
13:36
norm. Before the open AI event, we
13:38
had so many rumors that they were going to announce
13:40
a search product, which they of course didn't, but I
13:42
would be very surprised if something like it wasn't still
13:44
in the works. Still, maybe the
13:46
tweet of the day went to Andre Carpati,
13:48
formerly of open AI, who in reflecting on
13:50
the new open AI voice assistant, which reminded
13:53
so many people of Samantha from her, wrote,
13:55
the killer app of LLMs is Scarlett Johansson.
13:57
You all thought it was math or something.
14:00
AI is more competitive than ever, and next week
14:02
we're getting a Microsoft event as well, so stay
14:04
tuned for all of this, as there is going
14:06
to be a lot to discuss. If
14:09
you made it this far and want to dig
14:11
deeper into how to actually use these tools, please
14:13
go check out Super Intelligent. It's our platform for
14:15
fast, fun, and useful AI learning. You can find
14:17
it at vsuper.ai. For now, though, that's going to
14:19
do it for the AI Daily Brief. Until next
14:21
time, peace!
Podchaser is the ultimate destination for podcast data, search, and discovery. Learn More