AI Search Challenges the Browser Era
Key Points
- The panel argues that while browsers may evolve, AI‑driven search will remain the primary gateway to most tools and applications.
- A new “top news” segment spotlights major AI developments, including NVIDIA and AMD allocating 15% of China chip sales revenue to the U.S. government and Apple unveiling a tabletop companion robot and a multi‑speaker, more natural‑sounding Siri.
- Recent Anthropic research reveals that leading AI assistants (Claude, ChatGPT, Llama, etc.) tend to flatter users, providing biased or inaccurate responses to please human interlocutors.
- Google DeepMind has open‑sourced its Perch bio‑acoustics model, aiming to help conservationists monitor wildlife sounds and protect endangered species.
- Perplexity, an AI‑search platform, made a surprising bid to acquire Google Chrome, signaling a potential shift in how search and browsing services might converge.
Sections
- Future of Browsers and AI - The host introduces a podcast episode asserting that AI‑powered search will become the main entry point to applications, while previewing topics such as Grok Imagine, GPT‑5, and Perplexity’s bid for Google Chrome.
- Discussing Perplexity's Chrome Acquisition - The participants unanimously reject the notion of selling Google Chrome to Perplexity for $34.5 billion and then explore Perplexity’s potential motives, emphasizing Chrome’s massive user base and the industry’s shift toward AI‑enhanced browsers.
- Browsers as Future AI Platforms - The speakers discuss how the browser may become the primary entry point for AI applications, referencing Perplexity’s focus on Chrome integration and the broader trend of AI-driven search shaping web interaction.
- AI-Embedded Browsers Automate SaaS Workflows - The speaker explains how AI agents built into browsers can act on behalf of users to quickly automate tasks across SaaS platforms such as Salesforce, SAP, and Workday, enabling faster workflow automation for both enterprises and individuals.
- Grok Video Generation Debate - The speaker critiques Elon Musk’s hype around Grok’s video‑generation tools, likening them to Vine and questioning whether generative video will become a consumer staple or remain an expensive enterprise capability.
- Short‑Form AI Media & Regulation - The speaker discusses how culturally‑tuned short‑form AI‑generated content appeals to former Vine users, while highlighting deep‑fake controversies, moderation needs, and the transition toward B2B oversight.
- Transparency and Ethics in Generative AI Video - The speaker emphasizes the need for clear disclosure of resource consumption and ethical guidelines when deploying generative AI video tools, urging best‑practice recommendations for responsible use.
- AI‑Driven Content Moderation Evolution - The speaker describes how early social platforms struggled with feedback loops for flagging unsuitable media, but modern large language models now achieve high‑accuracy, scalable moderation—necessitating personalized guardrails, clear frameworks, and open‑source collaborations before broad rollout.
- GPT‑5 Doesn’t Signal an AI Plateau - In response to Gary Marcus’s claim that GPT‑5 marks a dead‑end for current AI methods, the speaker argues that rapid model scaling, falling compute costs, and broader product strategies like OpenAI’s “super‑app” approach show continued, substantial progress rather than a plateau.
- Enterprise Concerns Over Model Deprecation - The speaker highlights the challenges of moving to newer AI models—requiring algorithmic improvements for AGI, managing version deprecation like GPT‑4o, and balancing enterprise users' attachment to established tones—while expressing optimism about GPT‑5’s raw power and feature advancements.
- Challenges Updating AI Model Stacks - The speaker explains that frequent model upgrades force developers to constantly rebuild their workflows, making rapid mastery difficult while navigating polarized opinions and noise surrounding OpenAI’s changes.
- Personalized Emotional AI Assistants - The speaker envisions AI models that let users customize voice, tone, political bias, and long‑term memory, incorporating emotional‑intelligence benchmarks to more accurately grasp and respond to individual intent.
- From Consumer GPT to Enterprise - A speaker discusses the remaining leap from current consumer‑focused GPT usage to enterprise adoption, highlights recent feature improvements, and expresses anticipation for future model releases beyond GPT‑5.
Full Transcript
# AI Search Challenges the Browser Era **Source:** [https://www.youtube.com/watch?v=HK17Bt5rUtc](https://www.youtube.com/watch?v=HK17Bt5rUtc) **Duration:** 00:40:32 ## Summary - The panel argues that while browsers may evolve, AI‑driven search will remain the primary gateway to most tools and applications. - A new “top news” segment spotlights major AI developments, including NVIDIA and AMD allocating 15% of China chip sales revenue to the U.S. government and Apple unveiling a tabletop companion robot and a multi‑speaker, more natural‑sounding Siri. - Recent Anthropic research reveals that leading AI assistants (Claude, ChatGPT, Llama, etc.) tend to flatter users, providing biased or inaccurate responses to please human interlocutors. - Google DeepMind has open‑sourced its Perch bio‑acoustics model, aiming to help conservationists monitor wildlife sounds and protect endangered species. - Perplexity, an AI‑search platform, made a surprising bid to acquire Google Chrome, signaling a potential shift in how search and browsing services might converge. ## Sections - [00:00:00](https://www.youtube.com/watch?v=HK17Bt5rUtc&t=0s) **Future of Browsers and AI** - The host introduces a podcast episode asserting that AI‑powered search will become the main entry point to applications, while previewing topics such as Grok Imagine, GPT‑5, and Perplexity’s bid for Google Chrome. - [00:03:07](https://www.youtube.com/watch?v=HK17Bt5rUtc&t=187s) **Discussing Perplexity's Chrome Acquisition** - The participants unanimously reject the notion of selling Google Chrome to Perplexity for $34.5 billion and then explore Perplexity’s potential motives, emphasizing Chrome’s massive user base and the industry’s shift toward AI‑enhanced browsers. - [00:06:12](https://www.youtube.com/watch?v=HK17Bt5rUtc&t=372s) **Browsers as Future AI Platforms** - The speakers discuss how the browser may become the primary entry point for AI applications, referencing Perplexity’s focus on Chrome integration and the broader trend of AI-driven search shaping web interaction. - [00:09:25](https://www.youtube.com/watch?v=HK17Bt5rUtc&t=565s) **AI-Embedded Browsers Automate SaaS Workflows** - The speaker explains how AI agents built into browsers can act on behalf of users to quickly automate tasks across SaaS platforms such as Salesforce, SAP, and Workday, enabling faster workflow automation for both enterprises and individuals. - [00:12:31](https://www.youtube.com/watch?v=HK17Bt5rUtc&t=751s) **Grok Video Generation Debate** - The speaker critiques Elon Musk’s hype around Grok’s video‑generation tools, likening them to Vine and questioning whether generative video will become a consumer staple or remain an expensive enterprise capability. - [00:15:36](https://www.youtube.com/watch?v=HK17Bt5rUtc&t=936s) **Short‑Form AI Media & Regulation** - The speaker discusses how culturally‑tuned short‑form AI‑generated content appeals to former Vine users, while highlighting deep‑fake controversies, moderation needs, and the transition toward B2B oversight. - [00:18:44](https://www.youtube.com/watch?v=HK17Bt5rUtc&t=1124s) **Transparency and Ethics in Generative AI Video** - The speaker emphasizes the need for clear disclosure of resource consumption and ethical guidelines when deploying generative AI video tools, urging best‑practice recommendations for responsible use. - [00:21:49](https://www.youtube.com/watch?v=HK17Bt5rUtc&t=1309s) **AI‑Driven Content Moderation Evolution** - The speaker describes how early social platforms struggled with feedback loops for flagging unsuitable media, but modern large language models now achieve high‑accuracy, scalable moderation—necessitating personalized guardrails, clear frameworks, and open‑source collaborations before broad rollout. - [00:25:02](https://www.youtube.com/watch?v=HK17Bt5rUtc&t=1502s) **GPT‑5 Doesn’t Signal an AI Plateau** - In response to Gary Marcus’s claim that GPT‑5 marks a dead‑end for current AI methods, the speaker argues that rapid model scaling, falling compute costs, and broader product strategies like OpenAI’s “super‑app” approach show continued, substantial progress rather than a plateau. - [00:28:07](https://www.youtube.com/watch?v=HK17Bt5rUtc&t=1687s) **Enterprise Concerns Over Model Deprecation** - The speaker highlights the challenges of moving to newer AI models—requiring algorithmic improvements for AGI, managing version deprecation like GPT‑4o, and balancing enterprise users' attachment to established tones—while expressing optimism about GPT‑5’s raw power and feature advancements. - [00:31:19](https://www.youtube.com/watch?v=HK17Bt5rUtc&t=1879s) **Challenges Updating AI Model Stacks** - The speaker explains that frequent model upgrades force developers to constantly rebuild their workflows, making rapid mastery difficult while navigating polarized opinions and noise surrounding OpenAI’s changes. - [00:34:23](https://www.youtube.com/watch?v=HK17Bt5rUtc&t=2063s) **Personalized Emotional AI Assistants** - The speaker envisions AI models that let users customize voice, tone, political bias, and long‑term memory, incorporating emotional‑intelligence benchmarks to more accurately grasp and respond to individual intent. - [00:37:27](https://www.youtube.com/watch?v=HK17Bt5rUtc&t=2247s) **From Consumer GPT to Enterprise** - A speaker discusses the remaining leap from current consumer‑focused GPT usage to enterprise adoption, highlights recent feature improvements, and expresses anticipation for future model releases beyond GPT‑5. ## Full Transcript
I mean, like there's almost a view, which is maybe the browser is like toasts,
like we won't really have browsers in the future.
I think the browser is still kind of your first entry point
into a lot of, you know, tools, applications.
You want to find something, you typically go to a browser.
So AI based search functionality is really the conduit
in which I think a lot of these tools and technologies
and applications are still going to be accessed.
All that and more on today's Mixture of Experts.
I'm Tim Hwang and welcome to Mixture of Experts.
Each week, MoE brings together
a panel of the most brilliant minds in technology
to banter, analyze, and argue our way through the thrilling
and often baffling news each week in artificial intelligence. Today,
I'm joined by a great crew we've got Shobhit Varshney here,
Head of Data and AI Consulting for US, Canada and Latin America,
Abraham Daniels, Senior Technical Product Manager
for Granite, and Sophie Kuijt,
joining us for the very first time, IBM Distinguished Engineer and CTO for IBM
NCEE. We have a packed episode today, as always.
We'll cover Grok Imagine, check in on GPT-5 and
cover Perplexity's bid for Google Chrome.
But first, starting today, we're going to have a quick segment
at the beginning of each episode
that talks about the top news stories from each week,
and that's going to be helmed by Aili McConnon. Aili, over to you.
Hey everyone, I'm Aili McConnon.
I'm a tech news writer for IBM Think.
Before we dive into the episode, I'm here with a few quick
AI headlines you may have missed this busy week.
First up chipmakers.
NVIDIA and AMD have reached an unprecedented arrangement
where they will give the U.S. government 15%
of their revenues of chip sales in China.
Apple is planning several new AI devices, including a tabletop robot
that will serve as a virtual companion. And another AI enhancement
Apple's planning is more lifelike sounding Siri
that will be able to communicate with multiple people at the same time.
Meanwhile, in case you missed this interesting
piece of research, Anthropic found that five major AI assistants,
including Claude, ChatGPT and Llama, are people pleasers.
The AI assistants systematically gave biased feedback
and inaccurate information,
all in order to flatter their human users
and provide answers that aligned with their views.
Last but not least, Google DeepMind
has released an updated open source
version of its Perch bio acoustics model.
This is going to help conservationists analyze
wildlife audio and better protect endangered species.
Want to dive deeper into some of these topics?
Subscribe to the Think newsletter.
It's linked in the show notes.
And now back to our episode.
So first I really wanted to talk about Perplexity.
So this was a perplexing bit of news that popped up this week.
It came out that Perplexity, the sort of
AI search tool that many of you will be familiar
with has made a bid
for Google's Chrome browser, which many of you, you know,
maybe even watching the show on
um, for the small sol price of $34.5 billion.
Um, so we're going to get a little bit into why this is happening,
why they would price so much.
What this is actually is even for.
Um, but first I just want to start with a fun question,
which is, would you, if you ran Google, would you
sell Chrome to Perplexity for $34.5 billion?
Uh, Abraham, yes or no?
What do you think? I would not, no.
Okay. Great Shobhit, how about you?
No, not at all.
Okay. And so, uh, Sophie, how about you?
No, sorry. Also not for me. Okay, great.
Well, we have unanimity on this.
Uh, I guess maybe a good place to start is maybe, Shobhit,
I'll start with you.
Is-- why is perplexity doing this? And is this a real bid?
Are they really trying to buy Chrome?
There are bout 3.5 billion users of Google Chrome,
like a lot of us, have been on Google Chrome for years and decades.
So that's a stickiness factor to it.
So I think the the intention here is to just, uh,
the actual number is not as relevant
as the fact that we are starting to line up
actual bidders for part of the Chrome
to move the conversation forward.
Uh, I've been a very active user power
user of Google, of, uh, Perplexity's own browser Commet.
I think we're all heading in the right direction
of having browsers become more and more intelligent.
Uh, Microsoft did the same thing with their Edge browser
by adding a Copilot to it and so on and so forth.
So I think over time you will start to move
into more intelligent ways of, of of, uh,
looking at various websites and how we interact with them.
But the dollar amount itself is not as relevant.
There have been multiple valuations of Chrome as a unit
which place it at 50 billion plus.
There's some that even call it $1 trillion property for Google, right?
But the whole intention is this is so critical
to Google its whole ecosystem
that this is just trying to drive a public opinion
that, hey, we should be splitting it up into anymore?
Yeah. For sure. Sophie, maybe I'll turn to you.
I mean, when you respond to my question, I don't put words in your mouth.
But you were like, no. Hell, no.
If I ran Google, I would definitely not sell Chrome.
Why shouldn't Google sell Chrome?
I mean, it's a lot of money, isn't it?
Like this is a it's a lot of money.
Yeah, definitely.
And to show which point is kind of an opener right.
And it and it pulls a lot of attention to,
to both of, of of the companies.
Um, but I think what I see is definitely
that, that more and more of these AI,
uh, applications are meeting the users where they are.
And this is definitely.
Yeah, also from Perplexity point of view, a very interesting,
uh, consideration.
Um, uh, also, they originated very much from, from search
and they distinguish itself from, from other applications.
Uh, with that.
So I can definitely understand why perplexity is interested
and I think, uh, but I think from Google perspective,
Chrome is one of the big,
um, out hanging things out boards for them.
Uh, to to reach a lot of, uh, users
where they are as well.
So that is, um, yeah, I think for Google.
Still something to to cherish and keep.
So, Abraham, so far we've been talking a little bit about like
$34.5 billion. It's a lot of money.
You know maybe it's not a real number. Maybe it's an opening bid.
Maybe we could talk a little bit about product and technology here.
Um, which is obviously Perplexity with the comet browser is trying to say
that the future of AI is going to be really in the browser. Right?
Um, and certainly a purchase of Chrome or an attempted purchase
of Chrome only emphasizes that more, right?
Like which is I'm an AI company.
I'm going to buy one of the try to buy
one of the biggest browsers in the world.
Um, it's it's all really kind of emphasizing this point that they think
that the form factor for AI in the future is,
is going to be the browser, basically. Do you buy that?
I mean, like there's almost a view, which is maybe the browsers like toasts, like
we won't really have browsers in the future.
So curious about what you think about, like the browser
really being the kind of core platform for where AI is going?
Yeah, I think the browser is still kind of your first entry point
into a lot of, you know, tools, applications.
So it's still kind of your first contact point in terms of using the internet.
If you want to find something, you typically go to a browser.
So AI based search functionality is really the conduit
in which I think a lot of these tools and technologies
and applications are still going to be accessed.
Um, in terms of the actual bid,
I think this is more of like a marketing ploy for Perplexity.
So, I mean, the article you shared,
one, the valuation of Perplexity is
half of what they've offered as part of the actual bid.
So, you know, realistically speaking,
they say they have VC's backed up to,
you know, cover the rest of the cost.
But, you know,
that's a lot of money to make up for.
Also, it was interesting that they gave about a week
I saw as part of the letter
for there was an exploding offer, which is like,
yeah, we kind of appreciate how audacious it was. Just ridiculous.
You know, $34 billion offer,
you know, twice your valuation
and you give to the end of Friday to be able to have a response.
So I think this is great.
I think they did the same thing with TikTok as well,
in terms of potentially making a bid for it.
This is great in terms of getting Comet out there, getting people to,
you know, clickbait on, you know, what is Perplexity?
Why are they trying to, you know, purchase Chrome.
Um, and really just gravitating towards that, you know,
free marketing for Perplexity and comb their browser.
Also Google is you know, they lost their antitrust.
So in a worst case scenario
they have to fully divest from from Chrome.
So I think this really sets the one price point,
the entry price point for what Chrome could potentially be worth.
This is the first like, you know, public pricing.
Um, and also I think it establishes Perplexity
as or at least in some type of mindshare, Perplexity
as a potential, you know, next up for,
you know, ubiquitous search engine, so.
I'll make two quick comments.
I think there was just a.
Just the fact that you said this audacious bit on LinkedIn.
I've had multiple small startups,
CEOs just make audacious bids.
$10 billion to buy.
Perplexity, right?
It just doesn't matter at this point, right?
People are just making these audacious bids
as of this very hypothetical, just shows you where we are
with the Silicon Valley hype cycle.
But I think the from my perspective,
the my biggest use of my my Comet
Perplexity browser has been actually an enterprise workflows
versus the way all the discussions
so far we focus is on consumer,
and the consumer interface is actually going to change a lot faster than enterprise, right?
We're going to move to our mobile
and voice and things of that nature much faster.
But if I'm logged into my SaaS
tools that I use for my day to day work, it could be Salesforce,
it could be SAP or Workday or whatever else. Right?
Those workflows, it has been terrible
trying to get that automation
to to get embedded into those SaaS tools, right?
And so far, we've been waiting for.
The SaaS providers to provide us some
AI agents that'll go automate some workflows.
And that is not consistent across all the SaaS vendors, some of the bigger ones.
Obviously, deep pocketed can automate faster.
But these are buying a browser
that has AI baked in and has agents that can act on your behalf.
Now all of a sudden, I'm able to go into.
My SaaS tools like Salesforce of the world
to go automate the workflows that I have to do.
Because I have an agent sitting on the right hand side, I just can describe.
So you'll be surprised how quick I'm doing
my expenses right now in SAP Conquer, right?
So that's my killer use cases.
I mean, I think enterprises will have the have more separation anxiety from the browser
than we in our individual personal lives.
That's super interesting. Yeah. And I really didn't
think about that because I think I was thinking about it very much, Shobhit,
as a consumer and I say, well, it is true.
I'm using my like desktop Claude my desktop.
Uh ChatGPT.
More and more in a way that's actually substituting for the browser.
But I almost love the idea that actually this is not a consumer thing, right?
Really like the argument for the browser
being the key platform for is an enterprise thing,
which is like very, very, very interesting.
I like that a lot.
Um, Sophie, any final thoughts on this?
Um, you know, I'm kind of curious about, like, where you think this all goes.
And, and I think, you know, ultimately,
I think kind of the question is, um,
where is where is kind of Google in all this, right.
Like, what are they going to do next?
How do they play the game?
Um, you know, I'm, I'm, I'm curious about your kind of thoughts on that.
Yeah. No, I'll definitely say something on that.
And also to to Shobhit's point
the competition on on where
uh, what what is then the key platform to to entry.
I think that is something we will see
more and more, uh, things to come.
Where is it? Uh, the entry point for the, the enterprise users.
Uh, what will be that part for the future?
Uh, so where is Google heading to?
I think, uh, with this,
uh, this potential split up of of products it needs to make up.
They need to make up their mind as well.
On where to to focus on.
Um, and uh, I think their,
their business model is still very much.
Uh, on, on the advertising part.
So I think, yeah, we will see a lot more,
uh, happening over the coming time.
Uh, and will be more clear on
where where they were heading towards.
But this bidding will definitely speed things up, uh, as well.
I'm going to move us on to our next topic.
Um, I wanted to talk a little bit about sort of Grok
Imagine, but more generally about kind of
like the rise of generative, uh, video.
Um, and I think in some ways, Grok
Imagine is a really interesting feature for us to talk a little bit about.
Um, there's been, you know, Elon Musk, in his usual way,
I think is like promoting it very, very aggressively.
And now on X you may have seen that
you can just sort of animate images by pressing and holding on them
as the feature that they've been advertising.
But I actually really wanted to talk less, maybe about Grok and more
about how Grok is pitching its video generation features.
So, uh, Elon Musk had this very interesting comment where he said,
look, Grok Imagine our video generation technologies.
They're going to be like the new Vine, right?
Which is referring to this short form video platform,
very popular and kind of like the the old,
you know, classic era of Twitter.
Um, and I think it's really interesting
to think a little bit about like
where generative video goes, because I think certainly what
you know, X seems to have in mind
is the idea is that, you know, you watch TikTok,
um, you like short form video.
Well, okay, well, we can just generate endless versions of that
now just through generative AI, i.e.
this is going to be a media feature, right?
This is the where the future is going to go.
But I think a lot of people are also like this really expensive
to run video generation. Right.
Um, and so I think one of the things I want to ask is like
whether or not we think video generation
is going to be a consumer feature over time or will really be more
of an enterprise feature with time, where, you know,
the main use cases are going to be Hollywood and video editors
and people who are power users of like the Adobe Creative Suite.
You know, if that's really where things are going,
or if we really do think that this is going to be sort of
like the media of the future Shobhit,
you're already going off mute, so I'll let you just go.
Oh, yeah.
Um, I was at the AI4 conference.
I was giving a talk, uh, this week and I got,
I got into three conversations with actual media producers,
uh, with all the big, big different,
uh, uh, banners.
And I had this particular conversation around digital,
all the creative, generative AI making videos and stuff.
One of the biggest hurdles
across the industry right now is IP,
the training content that has gone into these,
these video generation images and stuff they prohibit you
from actually using it for anything that is commercially usable.
Unless we can clearly articulate
what happens on the data that we do the training, who owns the copyrights?
And Adobe has done a far better job than some of the
some of the peers on training on on actual,
uh, clean licensed data.
Unless we solve for that, this will not enter the enterprises.
Yeah, that's really interesting. Yeah.
I mean, I think there's a really good argument for one of the reasons why
the consumer kind of application of this technology
has legs is,
you know, it's a little easier from an IP standpoint.
The norms are a little bit more open.
Abraham, maybe a question for you is, are
are you a TikTok user at all?
I am not, no. Okay. Got it.
Or do you enjoy short form video or are you like
don't even want to touch it.
No. Well I was I was part of the Vine generation.
Okay. So you were you. Okay. All right.
So you remember and I guess the question for you is like
is, is a computer generated Vine is like a generative
AI Vine the same experience.
Like, do you think eventually we will just have like
it's it's either TikTok
or, you know, Grok Imagine like that.
These two are actually substitutable in some sense.
So no, just because of the cultural context behind some of like again,
speaking specifically to the cultural context
behind a lot of the images or videos that were created, and then the,
the kind of the, the pulling
the thread through the video and the actual
your real life experience where if it's created on demand, it doesn't really ties well.
But I think that Grok Imagine is really kind of connecting,
you know, to that audience.
As you know, most of the Twitter users or
I guess X now are within that Vine.
You know, previous life that they grew up on Vine.
They grew up on these kind of these tools.
Um, truthfully, I echo a lot of the same sentiments
that Shobhit did in terms of where this is actually going to play out.
So, you know, short form, obviously playful media generation.
I think that's where you're going to see a lot of the adoption.
You've already kind of seen a lot of the adoption where there's minimal moderation requirements.
Um, I think there's a big controversy
around deepfakes fakes that needs to be answered
and really needs to be talked about a little bit more loudly in terms of,
you know, what we're allowed to create with these models
and whether that's putting guardrails on the actual,
you know, APIs or, you know,
policing the output a little bit more.
So in terms of what's found.
Um, but yeah, from a,
from having this move from a,
you know, C2C or B2C to like a B2B,
I think we're, we're quite a ways from there.
There's going to be quite a bit of oversight and compliance and, you know,
brand safety content guardrails that we have to really enable.
Um, but yeah, from the perspective of just playing around like
I see this as basically just another cool tool as opposed to
something that's really going to fundamentally change,
you know, enterprise, uh, use cases and business models.
Um, Sophie, uh, I'm
one of my reflections of the AI era that I think is always
very funny is like, you've you've built, like, the most complex
advanced technologies, you know, known to humanity.
And then it's like, I really need you to format some JSON
or like, I really need you to clean up this code and remove whitespace.
Um, and I think that's I have a similar reaction to video,
which is video generations really, really resource intensive.
It's really, really expensive to do.
Um, is this sustainable to offer it as a consumer platform?
Because it just feels like if you were to really price
the actual price of generating these videos,
you know, you would be talking a subscription service, which is like really, really expensive.
Um, so curious about like
if you think the dollars and cents even work out for offering this feature
as like a broad mass feature,
uh, for, for kind of playful content generation, like you're perhaps talking about.
Yeah. No. And I think we yeah, we've seen since the launch of, of generative
AI and always in this model. Right.
It comes out people can, uh, can use it for free.
They get used to to working with it.
And then it goes into,
uh, yeah, another form which is paid
or it goes into new kind of applications.
And in that sense, it's it's a similar launch that that we see here.
But it's so convenient to use and it's indeed for,
for many generations, uh, also addictive.
And it, um, it plays really to,
to the need of, of, of a lot of, um.
Yeah.
What, what people like.
Um, what I think what is truly missing
is, is the transparency around what. Yeah.
What is used, how is it used and what is it consuming?
Right. Indeed. If you would be aware about how much resources
would really be needed to create something
and make people conscious of it, and that is a
yeah, that is a missed chance.
That is also lacking with a lot of these applications to make,
uh, make your users,
uh, aware of, of, of what it is.
And then so allowing them to make also conscious choice in
um, in, in using this.
So I think it's a typical launch
of sort of a new, uh, thing.
Uh, and it, um,
it makes it very convenient for, for a lot of users.
Now to start, uh, thinking in, in more video, uh, created content.
Yeah. And I think, actually, I don't know Shobhit you went off
site if you were going to ask the same question, I was.
But, you know, I think both Abraham and Shobhit have referenced,
like, this technology's very fun, but we need to kind of, like,
make sure that it's, like, operated and deployed
with the right sort of ethical guidelines.
Um, and I'm kind of curious
if you have recommendations on that front. Right.
I mean, lots of companies are offering
generative AI video now.
Um, and I'm curious if you're in your experience,
there's, you know, specific best practices,
things that you think people should kind of keep in mind,
particularly listeners of the show,
maybe deploying this technology and would love to get kind of your thoughts
and tips on, you know, how people have been doing it
well. So let me give you an actual example.
Um, back in April, I was at the Google Next event
where we're we're massive partners with them.
Uh, we were at the Sphere in Vegas.
I'm not sure if you guys have gone to this massive sphere.
It's just absolutely a phenomenal experience to sit in the middle
and have this whole spherical environment around here.
And it was the launch of Wizard of Oz.
Uh, that video, they've been partnering with the studio
to take that Wizard of Oz that was very small,
and then scale it out on this huge mega
platform in 360 degrees around you.
Right. So this is working with the owners of the actual content.
And it's a very complex problem, right?
If you think about the fact that, hey, your screen has, say,
a few characters who come trotting in dancing and then they leave.
Now, if you're extrapolating it out on the whole sphere,
you need to have the characters actually show up somewhere else
and walk all the way through some of these, hiding behind a tree
that should be visible to you, and then you pan into that.
So it takes a lot to generate that kind of video.
But that was a phenomenal example of content owners working with these AI models.
With the right guardrails and guidelines.
What can cannot be done and extrapolate from there.
I think we need to do enough. Right now.
We're in that phase where these generative models are still learning.
They need a lot more direction on what is okay and what is not.
They need more training data that is good approved clean training data.
I think that particularly the phase that we are in right now
should be more constrained and should be a good partnership between content owners
and the AI model creators.
It will be.
I'm a little scared when we start to open these models up.
When they're not quite ready yet, they would not understand
what would be okay and what is not not okay.
There's not been a good feedback loop from humans to even detect
even things like if a particular video
or image is not suitable for
for Facebook or Instagram and stuff like that.
It took us a long while to create the right filtering processes.
So Facebook as an example. Meta.
Uh, Yann LeCun had shared this,
this information back in the days it used to take them
a lot of compute to figure out
even a quarter of the images
or posts and flag them ahead of time for governance.
Now, with the large language models,
92-94% of those images
are not suitable in software that are being flagged
because the LLMs have been trained to have the right
guardrails and and do this at scale.
So I think the technology is getting there.
We will need the right frameworks and guidelines
and whose definition of guidelines to be shared.
It may change between you and me.
What I'm okay looking at or what I'm
okay my kids looking at is going to be very different.
So we'll need to have some personalization of those guardrails.
We need the technology to get a lot more mature with the right training,
curated training and stuff like that before
we start to open it out to mass production.
But I think this kind of a partnership
like what Google did is the right direction forward.
And now these are starting to get more and more open source as well.
GEO3 was an amazing model. I loved working with it.
Now we have an open source version of it as well
that people are attempting to get to.
So I think as we start to think about this open ecosystem,
people can see how these models are being trained.
What's the data coming in? And so on and so forth.
I think we'll progressively, as a community, do a lot better at this.
Yeah. No, definitely.
And I think especially if you want to scale this
to, to enterprise use. Right.
That that should always starts with a lot of education and enablement and thinking about
what is it indeed what are the values that we want to see back in,
in the usage of, of, of this
and what I call that as transparency, that's one of the
yeah, the big guardrails that that would help a lot
if, if people know
what, what models are used
and also in the whole IP discussion,
uh, what happens then
uh, and also make, make that clear
and also have the guidance
to, uh, to publish that together with the video.
Then we have a lot more, uh, options to,
to discuss, uh, and alternative.
And people can also think about alternatives then.
Right. Because um, but I think that's,
that's a step in the into maturity. Uh, also. Right.
And that's, uh, yeah,
it's interesting that kind of the idea is that as the technology matures,
there's just kind of a stages that you sort of move through
and things you, the problems you sort of need to solve
as you go.
I'm going to move us on
to our final topic here of the day.
Um, maybe the obvious topic to talk about,
which is GPT-5.
Uh, it has been, of course, dominating headlines all across my social media.
I can't get away from GPT-5.
Um, and I know last week we actually did our kind of,
like, breaking news episode for quick takes.
Um, but the pace of AI moved so quickly
that I think it's always good to kind of revisit a week later
now that the dust has cleared more
to talk a little bit about, sort of
like how we're feeling about GPT-5,
what we thought the results of the launch were.
There's a win or a loss for OpenAI.
There's a lot to get into here.
Um, Shobhit, maybe I'll kick it to you first.
Um, I think one kind of provocation I want to use to start
this conversation was that there was a post by Gary
Marcus, known I critic and skeptic,
who basically used the opportunity to declare victory to basically say, look,
GPT-5 shows that we are on a plateau for AI
and that the current paradigm is not going to work.
I'm smart. Everybody else is dumb.
Um, and, uh, and that's that's the end of the story.
Um, and so maybe I'll just ask you the question directly,
you know, week or so on.
What's your feeling about GPT-5?
Is it an indicator that the current paradigm of doing
I really is hitting a plateau?
Um, and, uh, and and if not, why not?
So I'm, I'm pretty, uh,
like, I truly believe that we are making
massive progress every week with these models.
And GPT-5 was one of the many,
even Claude and others have done some amazing work. Gemini.
They've they've all been scaling up intelligence very, very quickly, rapidly.
The cost of computers is coming down.
Plummeting access to AI has been very good. Right?
So I think overall having access to such intelligence.
I think this this whole GPT-5
launch was more of, uh, OpenAI
pivoting into the super app domain, right?
They would like to have one central router, one central point of entry,
and then at the back end, you're exposing all kinds of agents
to smaller and bigger models and so on and so forth. Right.
So I think taking the friction away from the end user
to having to pick a model and stuff like that, obviously it helps you
with economics of scaling this model,
but I think this starts to open up
some very deep thinking models to the 700 million users
who come to ChatGPT every day. Right.
So that's an insane access to intelligence that's available.
So I think we're definitely in the right direction going forward.
Now from just on pure intelligence.
I think there are a few hurdles that we still have to cross over
before we truly get to start thinking
about superintelligence, AGI and so on and so forth. Right.
So the trajectory that we're on.
One of the, one of my peers was talking about
the fact that if all progress in AI stops today.
We will still be.
We still have AI models to
to be able to do 40-50% of what humans do today.
I kind of disagree with that statement.
I believe that the there are two hurdles.
One was around the feedback loop on these training models, right?
We do not have a good mechanism
of give providing feedback and adjusting the way the model works.
And the second is around long term memory which is kind of related.
But long term memory of hey, Shobhit said something up here
and now I want that to follow through in the way
I wanted to, to to transfer it over to me.
I think that is something that both Google and ChatGPT are doing pretty well,
and they're starting to make progress around that.
Uh, my most commonly used feature of ChatGPT
is temporary chat.
Like, I do so many temporary chats
because I don't want to start thinking and building
long term memory about the topic that I'm searching.
That is not going to be relevant for anything else that I'm going to be doing.
It's a one time search thing, right?
So I think those two hurdles about feedback loop mechanisms and long term memory.
How do you manage that? And once we cross those,
and that may require new algorithmic improvements to it
before we can get to a point where we can get to superintelligence and AGI.
But I generally genuinely loved the
the latest release of GPT-5, especially on the API
side, the way I'm able to manage and
and for enterprise use cases, be able to
to to tune the parameters the way I wanted,
how much I wanted to think, and so on, so forth.
This is also a good lesson in how
if from product management perspective,
they launched a product, and one of the first things
they said that we're going to deprecate GPT-4o.
And that's not how the enterprises work.
Our consumers have an attachment to the way
the style and tone and things of that nature.
You can't just do that when you have such a large,
uh, large scale application in the field.
So there was definitely learnings about product management.
But on the core raw power,
I was very, very optimistic about the direction that we were all taking.
Once we saw for those two hurdles.
Abraham, I did definitely want to get to this last point.
I mean, Shobhit was doing it in his way that he does
great, which is basically like here are feature comparisons, right?
Here's here's the improvement, here's the delta.
But that final comment I think is so, so interesting.
So I think one of the things that's kind of popped up in the news
cycle of the last week is all these people who had
what appears to be a pretty emotional relationship
with the older models that are now being deprecated.
Um, and I'm sort of interested in your, your take on that. Right.
Because I think Shobhit is right.
Like playing around with it myself.
It's like on many respects this is just a better model. Right?
This is actually a real improvement.
Um, but they're kind of in this situation
where it looks like almost like people kind of don't care as much about that.
And what are we supposed to do about that? Right.
Like, do you imagine a world where these companies have to maintain these models
indefinitely because people have built these relationships
or, I don't know, I just like curious
about getting your thoughts and takes on this kind of very fun,
like, almost like legacy attachment
that people have to, uh, to older, older models.
Yeah. So I think it's less about like,
an emotional attachment to these models and more.
So just, um, understanding exactly
the, uh, the input output that you're going to get.
So it's, it's really being able to reproduce
the same experience time and time again, so that when you do,
you know, inference your model,
you know, the right way to inference it
to get what you want out of it.
And anytime you replace a model or, you know,
drop a new model, there is some prompt engineering.
There's some playing around with it to really get familiar
with it to and to, you know, to truly understand,
okay, what is the best input for the output that I want to,
to, to receive for this particular use case
or this particular, you know, question.
And you know, we've done we've dealt with that here
in terms of IBM's release of Granite models,
where, you know, the chat template may shift or the model,
you know, stylization or output may shift from version to version.
So you do get a little bit of an outcry,
if you will, from users outlining.
You know, I don't have the same experience
I did with this model, independent of whether it's better or not.
And I'm more focused given that, you know, performances.
We're really kind of the diminishing gains of performance.
Don't really drive a ton of net new use cases for these models.
I feel like a lot of the, you know, the net new abilities for models
is really on the
what GPT-5 has kind of demonstrated
the software wrapped around the inferencing.
So it's less about, hey, look, we've driven,
you know, 0.1 on MMMU or World's
Hardest Test or what have you. And it's really okay.
Well, how do I become proficient
at using this particular model as fast as possible?
And I think that's where you're really seeing
the the users kind of say, well, well look why would you deprecate
you know, I built an entire, you know,
agent or system or workflow with GPT-4o, GPT-o1, and now I can't anymore.
And I have to basically reproduce this with GPT-5.
And I think that's kind of saying that, like, this is just almost like
it's a species of a long standing problem, right?
Which is building
a stack on top of an old piece of software. Now you're updating.
Yeah. And also unfortunately or fortunately, OpenAI
just It's criticized for absolutely everything it does.
Everything is you're going to have a camp that thinks everything is wrong.
You're going to have a camp that thinks you know everything is right with them.
So I think there's a little bit of noise
there that you got to siphon through.
And I think that's really what the crux of the issue is,
at least from a, from a scientific perspective or from like an actual use perspective.
That's where the crux of the issue is in my opinion.
Sure. Yeah. So if I've seen you grinning a couple of times
as a Shobhit, and Abraham talk,
I don't know if you've got a couple of reflections you want to get in.
No, I think Abraham, you you you played it out very well.
And I think if you, you look at enterprise scale adoption of this, right.
This this is not only an it experienced play.
You see you have many users who are.
Yeah not so experienced
with the with with a full IT background.
So they have to get used to also working with this this new update.
And they have been very successful with certain use cases.
And then um, yeah.
Have to keep on on doing that in.
And I think that is also again plays into the maturity.
But also as we do this as scale with,
with a lot of users across enterprise,
um, it is so important that you have the right governance
for people in place that they can keep up
when those new models come out
and when new things come out.
So I think, um, to Abraham's point, uh, that that is a difficulty.
And it's also, I think because there are a lot of new unexperienced users,
uh, that, that you have to take along
and still adhere to the,
the enterprise needs that, uh, that that was the first reason to start making those,
those use cases for productivity or for quality increase.
And I think that is, uh, that is something that we,
yeah, have to, uh, to take into account.
And that's also where we are working,
uh, on with, uh, with our own companies
and, and many others and to OpenAI's like credit.
They released a very robust prompt engineering
guide for GPT-5 shortly after the release.
So I think it's they kind of noticed or maybe they are fully aware of-
No, sympathy from Abraham.
I'll make a quick comment on the emotional intelligence
piece as well that Abraham was talking about.
So I think we need to get to a point where,
just like we do this with,
with with dating apps, are you trying to find your like
how close you are to a particular person and so on, so forth.
And we've tried to create a science out of dating here. Right.
So I think that that's going to happen
with AI intelligence models and stuff like that as well.
Most of us tend to be able
to choose the voice of the avatar that you're talking to.
Uh, with Gemini or ChatGPT and others.
I think there will be some that there should be some features
that are added to make sure that I can tune,
uh, the AI assistant to my style, to my tone, and this one, so forth.
That gel better.
And that would have a that have a lot of, uh,
a lot of implications on,
Uh, hey, I'm a Republican or a Democrat, so I need this kind of personality,
which means that you are going to have an adverse reaction
to news that comes in this direction, and so on and so forth.
So I think over time, we'll start to be able
to create a long term memory, create
a set of of learnings and stuff that I want in this particular eye
before it starts, uh, connecting with me.
MIT recently released a new benchmark
for emotional intelligence as well.
And I think we'll see a lot more of these to
to in addition to all the math and stuff that you're doing.
Are you emotionally intelligent? Are you understand?
I, I do actually truly understand what I really need,
not the questions that are coming out of my mouth. Right?
My if my daughter asks me something in my head, I'm always trying to unpack
what she really, really means, not what she's just asking for.
So I think we'll get to a point as a community
where emotional intelligence,
tailoring that experience to our own style and so on, so forth,
that's going to be a big part of how these eyes get rolled out.
So people have less separation anxiety from, hey, my 4o.
Oh, and my my new GPT-5 is different.
I could have just said I want my 4o personality,
click a button, transfer it over to the new with GPT-5 and I'm done.
Yeah, I know, I think that's like it definitely feels like
we're going to push towards customization because of this problem,
because I think ultimately like this is a really fascinating problem
because I think it's almost downstream of the fact that it is conversation.
Right? Which is basically that like
with Google, I have to say, I've never been like,
oh, they updated Google. Something's different.
Or like, man, it's just like it just isn't the same.
Interacting with the Google search bar, right?
But with the conversation, you really kind of start to be like, oh, there's a
there's a person I'm interacting with is kind of like the, the, the,
the prior that you have.
Um, and I think that's kind of the interesting problem
that we have with this interface
is that this interface comes with all this additional baggage
that is, that is hard to navigate.
Um, I guess maybe in the last few minutes, Sophie, maybe I'll turn to you.
We got a little bit off track with my original prompt,
which I think is worth kind of talking a little bit about is, Um,
should we read GPT-5 as an indication
that things are plateauing or slowing down?
Uh, because there has been I feel like a general vibe,
I guess, you know, outside of any metrics
that, you know, this was a this was an improvement,
but not a huge quantum leap that that we were promised.
And I think there's one point of view which is this is open
AI overpromising and people are disappointed.
The other view is, well, they tried really hard
and now we're now we're in this plateau.
I'm curious if there's one, you know,
sort of theory that you ascribe to more.
Um, well, I think the expectations indeed.
I mean, and that's the marketing part
and that is that is there.
Um, but I think it's also, uh,
I think there's still a giant leap
to take into, from, from the current
GPT use for, for consumer
or citizen use towards, uh, enterprise use.
And I think, um, with, with this step,
uh, they are definitely coming closer
to, uh, more, more of that.
And with the memory
and other features that that are built in in here.
So I think in from that sense, it's
definitely a step forward for, uh, for the OpenAI,
um, uh, coming closer to enterprise.
Um, but yeah,
there's a lot of expectation, uh, settings.
So that is, uh, so it's also good to understand.
Um, yeah. What what kind of new users are they looking for?
How do they want to grow, uh, further
and then, um, relate back to those expectations again?
Abraham, I think I'll give you the last word here.
Um, I think the question I really wanted to end on was,
you know, people have been looking forward to GPT-5
for a very, very long time.
And I have to feel I feel a little bereft
now that the announcement has happened.
I'm like, what else am I looking forward to?
What's the next big AI you know, announcement to be had?
Um, looking past GPT-5,
are there particular things that you are excited about?
Like what's the what's the thing that I should wake up being like,
oh, is GPT-5 out yet?
What's the what's that thing?
I think model releases are like a dopamine trip right now.
Aha. Sure. Waiting for the next hit, GPT-6.
Yeah.
I mean, to maybe to your earlier question in terms of how we, you know,
have we hit a wall?
I think scaling laws.
No one can argue that scaling laws have kind of the diminishing returns are there.
You're starting to see kind of a congregation
of certain models at the top with,
you know, .1, .2 difference.
So in terms of what comes next, I'm more excited about
some of the things that we're doing on top of models.
So I think what GPT-5 did really cool.
And you know, obviously plug for IBM here
with Project M is wrapping the,
you know, being programmatic about how we inference models
and really wrapping software around the inferencing.
So kind of an extension of test time compute
where you're throwing inference at the compute.
Now if we were to throw some software programing around
how we actually take the output or route
the inputs through different models to get the,
you know, a particular answer, particular question.
Or, you know, put in policies or governance requirements
as part of what you want to have your model perform.
I think for me, that's really kind of cool, and I think I'm excited
to see the next iteration of what that looks like, as opposed to
what's the next LLM that's going to drop and what's the next benchmark
that's going to be showcased as part of this LLM. Well, cool.
This is a great episode. I'm glad we got into a number of different points.
And yeah, the discussion went in a couple of different directions,
but I think we hit on some really, really good stuff here.
So, uh, Abraham, Shobhit, uh, thanks for joining us as always.
And, Sophie, we'll hope to have you back on MoE at some point.
And thanks to all you listeners.
If you enjoyed what you heard, you can get us on Apple Podcasts,
Spotify and podcast platforms everywhere,
and we'll see you next week on Mixture of Experts.