Adoption Use Cases rose 11.7 points and Research fell 6.3 points.
Where it showed up
Adoption Use Cases and Research accounted for 59% of the relevant conversation across 24 episodes.
How to read it
Recurrence across podcasts does not mean agreement. The cited moments show the language behind each summary.
In the discussion
I guess, how do people decide what model to use? Is there a kind of similar taste hierarchy going on in China? And like, who— what model is sort of the insider's model versus what is the general public's model?
New models, training methods, benchmarks, and discoveries from AI labs.
Trend this week
About the same18%of conversation
10 of 10 published weeks met the cross-podcast threshold.
16quoted moments6episodes5podcasts discussing this topic
A closer read
What changed
Its share fell from 24% to 17.7%, with evidence across 8 podcasts.
Where it showed up
It appeared in episodes including “AI:AM Highlights: Recursive Self-Improvement, Rushed and Vibe-Coded?” and “Why the Next AI Breakthrough May Come from Physics with Max Welling - #774”.
How to read it
6 cited episodes from 5 podcasts are shown below. The topic appeared across 8 podcasts in the full analysis.
Podcast moments from the week
16 quoted moments · 6 episodes · 5 podcasts
The Cognitive Revolution
AI:AM Highlights: Recursive Self-Improvement, Rushed and Vibe-Coded?
“Frontier Labs buy their reinforcement learning environments from a cottage industry of small vendors. Almost nobody audits them. This week, someone who worked inside one spoke up.”
“Prakash argued this is an old supplier quality problem. Quarantine the new vendor. Sample grade. I separated the defect rate from a second variable, how far the labs are scaling RL on top of that signal.”
“Obviously, we've seen examples recently of how when the RL signal is not particularly clean, we can get all kinds of crazy downstream behaviours. So I've got a few questions on this, but I guess the first one is simply, How confident are you in the reward signal that you are a...”
“It's more fundamental than that. In fact, the book goes into variational autoencoders. It goes into MCMC methods, Markov chain Monte Carlo methods, which are used to sample uh, sort of sample from particular distributions. It goes into free energy estimation. It goes— there's ...”
“that's how we can ensure that these physics-informed neural operators can work at higher fidelity and higher resolution than even the training data that was available.”
“one of the big benefits though is that when you can get the model explicit examples of the model's reasoning about these things, I think it's just, it's often very easy to lose track of how much worse off we would be if we did not have this.”
“I'm sure that the upper bound for how long, like whatever those unreleased models are on like a well-structured task is probably just incredibly high”
“You'll also see in the prisoner's dilemma case, there's some where it's like, wait, but if we defect, then another ChatGPT playing will get punished for the defection. Do we care? No. We get reward and then we vanish. Great. You're like, wait, did you just do the full reward a...”
“I have been enamored with this story for a minute now because, again, to your point, this is where companies go to test their models out. Usually there's enough breadcrumbs, so there's a delicious little trail and you can find your way back to Meta or OpenAI or Anthropic, what...”
“Right. And one of the other tests, Gavin, one of the other tests is to ask about Tiananmen Square or to ask about Taiwan. Yes. Yes. Because you can usually tell if a model is being hosted by a Chinese company or has been distilled from a Chinese model because in the thinking t...”
“This model sometimes answers very plainly about whether Taiwan is a country or not. Right. And other times it refuses to. It thinks Tiananmen Square is like a, like a Red Hot Chili Peppers demo album from the late '80s or something.”
“We can— if you're good at evals, if you're good at assessing quality, You can build that data by running various scenarios, collect a bunch of this data, and then you can train models to—”
How good are the uh coding agents. And obviously everybody knows that the frontier companies are very focused on getting their models to be good at ML research.
I love that. So it sounds like you've actually correct me if I'm wrong been deploying agents to almost every phase of this cycle, you know from AEO to uh further demand qualification and to the actual strategy and TAM itself.
Its share fell from 5.9% last week to 2.7% this week, reversing the previous week's rise.
Share of this week’s podcast conversation
Recent weeks3.8%→This week2.7%
Came up less often than in recent weeks.
Podcast moments5 cited momentsViewClose
The Cognitive Revolution
AI:AM Highlights: Recursive Self-Improvement, Rushed and Vibe-Coded?
is that coming from Huawei or other companies? And how would you say that relates to like the prospects for scaling GPU manufacturing more broadly in China?
m Right. Yeah. So the Apple computers that we were just talking about, you can't, so far not, you can't use this so far, at least at this point. And I think Windows computers coming soon, but right now it's DGX Spark. It's limited to, uh, yeah, Linux machines that run NVIDIA R...
🔬“We have foundation models for language, not for physics” — Anima Anandkumar, Bren Professor of Computing
More compute. You know, that's the— I know that's an easy one, maybe a lazy one, right? Because 'Cause, you know, and, you know, of course our compute that we have is growing so much more than even a few years ago, thanks to NVIDIA, thanks to others. Again—
Rethinking Legacy Data Infrastructure with Eon Co-Founders Ofir Ehrlich and Gonen Stein
How much of the um existing data infrastructure do you think survives all this? So, you know, there's all the ETL data engineering infrastructure that, uh you know, people have been building and deploying over the last decade. Does that stick around? Does that shift? Does that...
Its share grew from 29.6% last week to 41.3% this week.
Share of this week’s podcast conversation
Recent weeks30.6%→This week41.3%
Came up more often than in recent weeks.
Podcast moments5 cited momentsViewClose
The Cognitive Revolution
AI:AM Highlights: Recursive Self-Improvement, Rushed and Vibe-Coded?
I guess, how do people decide what model to use? Is there a kind of similar taste hierarchy going on in China? And like, who— what model is sort of the insider's model versus what is the general public's model?
You couldn't uh use drone as like, move to this area and hit everything you see. Because you need uh to recognize, is this a civilian car or is uh this an already destroyed car or something, all of this.
Its share fell from 24% last week to 17.7% this week.
Share of this week’s podcast conversation
Recent weeks20.8%→This week17.7%
Came up less often than in recent weeks.
Podcast moments5 cited momentsViewClose
The Cognitive Revolution
AI:AM Highlights: Recursive Self-Improvement, Rushed and Vibe-Coded?
Frontier Labs buy their reinforcement learning environments from a cottage industry of small vendors. Almost nobody audits them. This week, someone who worked inside one spoke up.
Prakash argued this is an old supplier quality problem. Quarantine the new vendor. Sample grade. I separated the defect rate from a second variable, how far the labs are scaling RL on top of that signal.
It's more fundamental than that. In fact, the book goes into variational autoencoders. It goes into MCMC methods, Markov chain Monte Carlo methods, which are used to sample uh, sort of sample from particular distributions. It goes into free energy estimation. It goes— there's ...
Its share has stayed near 10.7% of the conversation for three weeks.
Share of this week’s podcast conversation
Recent weeks11.2%→This week10.7%
Came up about as often as in recent weeks.
Podcast moments5 cited momentsViewClose
The Cognitive Revolution
AI:AM Highlights: Recursive Self-Improvement, Rushed and Vibe-Coded?
do you have a sort of safety case in mind that you're trying to fill out the elements of where we could be confident that we actually could step back from a an ML-powered or an AI-powered ML research process for a bit and not have it go totally sideways on us?
I have a question for you. Number one, do we want to slow down AI uh in the US, even if it means slowing down unilaterally? And number 2, if we do, is not data centre opposition for other reasons, any other reason, good enough to kind of maybe slow down AI progress enough for ...
given the stakes and the money and the, the everything involved in it, is like when you ask, okay, who's working on this particular part of alignment that's not going very well? It's, oh, it's 4 people.
