The AI Daily Brief: Artificial Intelligence News and Analysis - The 7 Most Important Things We Learned About AI This Week
Episode Date: November 23, 2025This episode breaks down the 7 most important things this week revealed about AI: Google’s return as a serious contender, fresh evidence that pre-training still has room to run, how shared infrastru...cture advantages are starting to compound, why multimodal and multimodal reasoning are only just getting started, why coding remains the most strategic battleground, and what it means that even Nvidia’s blowout earnings can’t fully support current AI market narratives.Brought to you by:KPMG – Discover how AI is transforming possibility into reality. Tune into the new KPMG 'You Can with AI' podcast and unlock insights that will inform smarter decisions inside your enterprise. Listen now and start shaping your future with every episode. https://www.kpmg.us/AIpodcastsRovo - Unleash the potential of your team with AI-powered Search, Chat and Agents - https://rovo.com/AssemblyAI - The best way to build Voice AI apps - https://www.assemblyai.com/briefLandfallIP - AI to Navigate the Patent Process - https://landfallip.com/Blitzy.com - Go to https://blitzy.com/ to build enterprise software in days, not months Robots & Pencils - Cloud-native AI solutions that power results https://robotsandpencils.com/The Agent Readiness Audit from Superintelligent - Go to https://besuper.ai/ to request your company's agent readiness score.The AI Daily Brief helps you understand the most important news and discussions in AI. Subscribe to the podcast version of The AI Daily Brief wherever you listen: https://pod.link/1680633614Interested in sponsoring the show? sponsors@aidailybrief.ai
Transcript
Discussion (0)
Today we are talking about the seven most important things we learned about AI this week.
The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI.
All right, friends, quick announcements before we dive in.
First of all, thank you to today's sponsors, Blitzy, Rovo, Landfall, IP, and robots and pencils.
To get an ad-free version of the show, go to patreon.com slash AI Daily Brief,
or you can subscribe on Apple Podcasts.
If you are interested in sponsoring the show or really learning anything else about it,
Visit us at AIDailyBrief.A.I.
Or send us a note at sponsors at AIDailybrief.
Lastly, it is the last couple of days of our AI ROI benchmarking study.
Get in your use cases by Tuesday to get the full report in a few weeks when it is ready.
All right, friends.
Well, welcome back to another weekend episode, which means it's a big think, long reads type of episode.
But as I sometimes do after a big week or a couple of weeks, instead of reading something
else or even basing this off of someone else's thoughts, I'm just going to get a little bit
extemporaneous about what I think are the things that we learned this week about the AI space,
in particular what I think are the most important things we learned about the AI space.
Now, there's a lot more, but these are the seven things that stood out to me when I was driving
home from New York last night, thinking about all of the implications of really not just the last week,
but the last couple of weeks. I think that we will look back on this couple week period as
wildly significant, both on a very personal and professional level in terms of the capabilities
increase that all of us now have access to, but also in terms of the dynamics in the AI race.
And that is where we will start. First, most important thing that we learned about AI this week,
Google is a player, and Sam Altman and OpenAI are worried. Now, obviously, Google was a player
before this, but their return to the top of the heap has been something to watch. It wasn't all
that long ago that they were caught totally behind and kind of embarrassed by OpenAI in the launch
of ChatGBTBT, only to launch a very dubious product in Bard,
which then gave way to the first rushed versions of Gemini,
which had all sorts of problems,
including wild recommendations in AI overviews and AI search,
as well as some very questionable choices
in terms of historical accuracy when it came to image generation.
And so you had like an 18-month period there
where that was what people were thinking of
when they thought about Google and AI.
Things started to shift, of course,
with the release of Notebook L.M.
For the first time and a long time,
there was an AI consumer product
that people really genuinely loved.
Now, specifically, it was the audio overviews feature that really captured people's attention,
but it turned out that it wasn't just the novelty of the audio overviews.
The entire suite was really useful.
And to the extent that audio overviews were the thing that got people into notebook,
they stuck around for a variety of other features which have continued to evolve.
That's kind of where we were heading into this year.
Now, the 2.5 series of models were really good.
Flash was incredibly useful from both a speed and a cost perspective,
and pro contended with the other models at the top of the pile,
on a lot of different types of use cases. Obviously, however, the launch of Gemini 3 and the
companion launch of Nanopro has really put Google into the stratosphere and completed this three-year
return-to-form journey that they've been on. One interesting thing that was dug up by the
information this week was that in advance of Gemini 3, OpenAI boss Sam Altman had actually
sent a memo to his team about what he anticipated to be rough seas ahead. According to the information,
open AI researchers had discovered or heard that Google had created new AI that had, in their words,
leapfrogged open AIs in the way that it was developed.
Altman said that their recent progress in AI could, quote, create some temporary economic
headwinds for our company.
He said, we know we have some work to do, but we're catching up fast.
And he cautioned employees that he, quote, expected the vibes out there to be rough for a bit.
Now, the broader story here is the competition coming from all sides for Open A.
right now. As the information points out, Anthropic has done a tremendous job this year,
increasing their revenue from developer-focused use cases as well as their API. You've got
Google surging even before the release of Gemini 3 and Nanobanana Pro with their Gemini
app, reaching number two and even at one point beating out ChatGPT as the top free app and
hitting 650 million monthly users. Now, in that memo, Altman recognized that OpenAI still does
have a brand advantage. He said ChatGPT is AI to most people and I expect that to.
continue. But it's no doubt that the company is heading into a more difficult period.
Now, one thing that is positive for the field as a whole, even if it does put competitive pressure
on OpenAI, is what the launch of Gemini 3 suggests about pre-training and scaling laws.
In short, the argument that we've hit a performance plateau or a wall looks a lot more dubious
today than it did about a week ago. After Gemini 3 was released and shared all of its impressive
benchmarks, including a few that saw just incredibly big jumps, such as its screen understanding
benchmark, which more than doubled the previous state of the art, Google Deep Mind's Oriole Vignols
writes, The secret behind Gemini 3 is simple, improving pre-training and post-training. Contra,
the popular belief that scaling is over, the team delivered a drastic jump. The delta between
2.5 and 3-0 is as big as we've ever seen. No walls in sight. Now, after OpenAI responded
to the launch of Gemini 3 with GPD-51 Codex Max and GPD 51 Pro, their researcher,
Nome Brown said something similar. He wrote,
Today we are releasing GBT-51 Codex Max, which can work autonomously for more than a day
over millions of tokens. Pre-training hasn't hit a wall, and neither has test time compute.
Now, Oriel actually was talking about this as well, that it wasn't just pre-training,
but also post-training and all the strategies that we have after the model has been trained
to get more performance out of it. Indeed, Oriel called post-training a total greenfield.
He said, there's lots of room for algorithmic progress and improvement, and 3-0 hasn't been an
exception. Now, this is all good news for a number of reasons. Alton seemed to acknowledge this in that
note, saying at one point by all accounts, Google has been doing excellent work recently. The information
points out, Google's success with pre-training in particular came as a surprise to many AI researchers,
given that OpenAI at times has struggled to eke out gains from pre-training, an issue Google also
wrestled with for a while. Apparently, by the way, OpenAI has a new LLM that is codenamed
shallop Pete that takes a different approach to pre-training and fixes bugs that they had previously
encountered. Still, moving back to the implications of the models that were released this week,
not only is it good news for consumers that there's more gains to be had, it's also good news for
investors who are betting on the AI theme. One of the biggest things that AI bears bring up
is the potential that we run into these sort of performance plateaus and walls that ultimately
also lead to a plateau in demand below the rate where it would sustain all of these big
infrastructure buildout deals that have been signed in the recent months. And in fact, it seems
like there is still room to run is genuinely a good thing for basically everyone in this
and all the consumers who are using these tools.
Now, moving away from just Gemini 3 strictly into Nanobanana Pro as well, but abstracting
it a little bit, it does feel like you're sort of starting to see the resource advantage
that Google has show up.
The information again points out the disparity.
They wrote, OpenAI is one of the fastest growing businesses in history, going from next to
no revenue in 2022 to a projected $13 billion this year.
By the way, Sam Altman says that that's actually closer to $20 billion.
They continue, but it is a lot of money.
also projected it would burn more than $100 billion in pursuit of human-level AI in the coming
years while spending hundreds of billions of dollars to rent servers to do it, meaning it will
likely need to raise the same amount in additional capital. Meanwhile, Google, valued at $3.5 trillion,
generated more than $70 billion in free cash flow over the past four quarters alone.
While ChatGPT looks poised to take a bite out of Google's search, Google's financial performance
has improved. In parts because it also has a booming cloud business that rents out servers
to large customers, including OpenAI and Anthropic.
Financial disparity between OpenAI and established firms like Google has prompted
public market investors to question whether the startup's unprecedented revenue growth,
including projected growth, will be enough to erase concerns about its future cash burn.
Now, hold aside whatever the market is thinking about this, because frankly, I care a lot less
about that.
I think where you're seeing the resource advantage show up is in and around multimodal.
Google is not just flexing with their core model.
they're also flexing the things around it.
We haven't had an update about an open AI image generation model for months and months and months,
unless you consider SORA 2 as part of that,
whereas Nanobanana and now Nanobanana Pro are out here really, really transforming
what it seems like is possible with image generation.
The reason that Google is able to do multiple things at once is that resource advantage,
and I wonder how that's going to start to create more and more distance in space between
them and competitors.
Remember, Anthropic at some point decided not to take.
even compete on that multimodal dimension. And I think the question that some will ask is,
will Open AI have to make similar types of decisions? This episode is brought to you by Blitzy,
the Enterprise Autonomous Software Development Platform with infinite code context. Blitzy uses thousands
of specialized AI agents that think for hours to understand Enterprise-scale code bases with millions
of lines of code. Enterprise engineering leaders start every development sprint with the Blitzy
platform, bringing in their development requirements. The Blitzy platform provides a plan,
then generates and pre-compiles code for each task.
Blitzy delivers 80% plus of the development work autonomously
while providing a guide for the final 20% of human development work
required to complete the sprint.
Public companies are achieving a 5x engineering velocity increase
when incorporating Blitzy as their pre-IDE development tool,
pairing it with their coding pilot of choice to bring an AI-native SDLC into their org.
Visit blitzy.com and press get a demo
to learn how Blitzie transforms your SDLC from AI-assisted to AI Native.
Meet Rovo, your AI-powered teammate.
Rovo unleashes the potential of your team with AI-powered search, chat, and agents, or build
your own agent with Studio.
Rovo is powered by your organization's knowledge and lives on Atlassian's trusted and secure
platform, so it's always working in the context of your work.
Connect Robo to your favorite SaaS app, so no knowledge gets left behind.
Rovo runs on the teamwork graph, Atlassian's intelligence layer that unifies data across all of
your apps and delivers personalized AI insights from day one. Robo is already built into
Gira, Confluence and Gira service management standard, premium, and enterprise subscriptions.
Know the feeling when AI turns from tool to teammate? If you rovo, you know. Discover
Rovo, your new AI teammate powered by Atlassian. Get started at ROV as in Victory, oh.com.
If you're listening to this, you already know how fast AI is writing the rules for innovation,
disruption, and value creation.
And this new era demands a new kind of patent law firm.
Landfall IP was built from the ground up to operate differently,
orchestrating how human expertise and AI work together for better patents at founder's
speed.
Created by world-class patent attorneys who saw a better way,
landfall IP lets AI execute the repeatable while attorneys elevate to create the exceptional.
Landfall isn't adapting to AI.
They were built for it.
Have a new idea?
Try the discovery agent for free.
It's a confidential tool that helps innovators synthesize their inventions,
and instantly see patentable insight.
Visit landfall IP.com to learn more.
That's landfall IP.com.
Today's episode is brought to you by robots and pencils.
When competitive advantage lasts mere moments,
speed to value wins the AI race.
While big consultancies bury progress under layers of process,
robots and pencils builds impact at AI speed.
They partner with clients to enhance human potential through AI,
modernizing apps, strengthening data pipelines,
and accelerating cloud transformation.
With AWS certified teams across U.S., Canada, Europe, and Latin America,
clients get local expertise and global scale.
And with a laser focus on real outcomes,
their solutions help organizers work smarter and serve customers better.
They're your nimble, high-service alternative to big integrators.
Turn your AI vision into value fast.
Stay ahead with a partner built for progress.
Partner with Robots and Pencils at Robots and Pencils.com slash AI Daily Brief.
Now moving on from the big competitive in AI race dynamics,
just to the new capability set from this week,
one of the things that was extremely clear
interacting with Nanobanana Pro
is that it really feels like we've barely scratched the surface
on what native multimodal AI can be.
You know, one of the interesting commentaries
following the launch of GPT5
was from Altman once again,
basically saying that in some ways
there was only so much more performance
that they could eke out from LLM and tech-based chat,
but there was still a ton of opportunity
across other AI modalities.
It really feels like to me that the way that the Gemini 3 suite, including Nanobanana Pro,
integrates the native reasoning of Gemini 3 plus the image generation capabilities of NBPro,
shows a glimpse of what a native multimodal future can do.
When you ask Nanobanana 3 to create an infographic,
it's not just that it does a good job on the visuals,
or even that it does a good job rendering the text, although it does.
It's that it's able to understand the source material
and do the work to consolidate and compress the information,
making judgments about what it should and shouldn't share,
and all of that informs the ultimate output,
which is the visual infographic.
Now, that's just one tiny use
that shows the different type of capabilities
that will exist in a natively multimodal regime,
and that's the sort of thing we have to look forward to.
Speaking of which,
turns out that reasoning plus text and images
opens just an absolutely insane number of use cases.
If you have not yet listened to Friday's episode about the 25 new things you can do
with Nanobanana Pro and Image Generation that you couldn't just a little while ago,
you really should go check it out just for the sake of all of the inspiration that you're going to get.
I have this concept of a utility score, which is basically a way of looking at new models
in terms of not what they hit on the standard academic and industry benchmarks,
but instead, how many new things we can do with them that weren't possible before.
And this week just smashed open a lot of those barriers,
The way that we share visual information is going to change.
The way that we study and educate is going to change.
I just experimented with doing infographics as a standard part of releasing my episodes.
It very much feels like we are at the beginning of a new journey
when it comes to discovering the use cases that these new capabilities open up.
Now, moving away from Nanobanana for a minute,
it is also clear after this week that while we might be distracted a little bit
with these flashy new visual capabilities,
coding is and remains a key battle graph.
for especially professional AI.
Now, part of that is that one area where Gemini 3 wasn't completely dominant instantly on
the benchmarks was around coding.
In fact, Gemini 3 Pro was behind Codd Sonnet 4.5 and GBT 51 when it came to SweetBench
Verified, not far behind, but a little.
More than that, OpenAI's big response to Gemini 3 was actually not even 51 Pro,
which only got a tweet announcement.
It was instead this new coding model, Codex Max.
When Sean Wang, the host of latent space and the curator of the AI Engineer Summit,
better known as SWIX, announced that he was moving to cognition.
Part of the reason that he gave is that he thinks that code AGI is about 80% of the rest of
AGI, and so why not work on that now?
And you get the sense that a lot of the labs agree with him, at least in terms of the
significance of that particular area.
Now, of course, it is notable that the outputs of coding may also get a benefit from other
parts of the developments this week.
I'm thinking in particular about vibe coding platform Replit's new design mode, which is powered by
Gemini 3, which significantly ups the level of visual quality and design of vibe coded projects.
And so all of these things are to some extent connected.
Still, I think that while we didn't anticipate just how central to the entire 2025 AI story
coding was going to be, I anticipate that it will be every bit as central in 2026, if not this time
unexpectedly.
Lastly today, we have to talk about the markets.
there was a brief moment, long enough for me to get a part of an episode out, where it looked like
the NVIDIA blowout earnings report and projections, had temporarily at least popped the AI
bubble bubble. Jensen Huang reframed the whole AI bubble conversation talking about the three
paradigm shifts happening simultaneously, and initially markets bought it. They surged. The next day,
however, Invidia was down again, and it's very clear that right now, the market is just not
comfortable with where it is. Now, I tend to think that there's a lot more going on than just an
AI. I think that AI-specific factors are part of the story. I think that the 1.4 trillion of deals
that Open AI announced was just a little bit too much for the markets to digest comfortably
and actually increase the overall level of skepticism. But I also think that the markets have
pinned their entire hopes and dreams on AI for the last three years, ever since the cutting cycle
began. And there are just too many other things that aren't going all that well outside of AI that are
weighing on the hole. We don't have any real economic data for the last couple of months because of the
shutdown. We have an extremely volatile political economic environment. We don't have any clarity
around what the Fed is going to do when it comes to monetary policy. At the time I was prepping this
episode, the Fear and Greed Index was down at something like eight, just incredibly fearful. And so like I
said, well, I do think certain parts of what's going on are AI-specific. I also think that there is a
much bigger picture that for the first time in a very long time, even AI isn't able to sweep under
the rug. Still, while that's the case, I do notice a bit of an increasing sophistication around
the market discourse on AI in ways that I think could be really positive over time. Gavin Baker,
who is at Gavin S. Baker on Twitter slash XAI, wrote a great piece that's pinned to the top of
his profile called some thoughts on AI, where he argued that Gemini 3 was the most important
AI data point since the release of 01 because of the way that it showed scaling laws for
pre-training are intact. Now, his piece goes into a lot of the economics around chips,
residual value in GPUs, ROI of AI, and comes to the conclusion,
all of this suggests we are still very early in AI.
I understand the OpenAI jitters.
The one trillion of unfunded spending commitments cast unfortunate doubt on the powerful
underlying reality of AI today.
OpenAI has lost share and is decisively behind two other companies from a model quality
perspective for the first time.
However, as Gavin points out, the internet trade survived the demise of Yahoo, MySpace, and
AOL. I don't think OpenAI losing share to Google and or others will materially impact overall
token demand and token demand as a function of customer ROI is what ultimately matters.
The share of those tokens will matter to the relative market caps of Google, OpenAI, XAI, and
Anthropic, but overall, token demand is what will drive all of the suppliers.
Ultimately, he concluded, tonight will be just one data point in what I think will be a decade
of steady AI progress. And on that note, the thing that I want to close with, bringing it back
to us personally is that if there is one key thing to take away from this week, is that more
so than basically any other week in 2025, you can do way more right now with AI than you could
a week ago. This has been, by a mile, the most spectacular capability increase period we have
had for an extraordinarily long time. We are barely scratching the surface on what we can do
with all these new tools and toys, and I cannot wait to get back to trying them out.
So with that said, I will wrap it here.
Appreciate you guys listening or watching as always.
Until next time, peace.