Dwarkesh Podcast

Our reading — their words are on the recording

0:00:46
Our reading
Between thirty and forty per cent of the Exploit Gym tasks these agents were set were unintentionally impossible to solve.
Their wordsBut in many of these cases, this vulnerability is just not broad or deep enough to ever actually be exploitable to get the flag. So a bunch of exploit gym problems are just unintentionally impossible. The authors estimate roughly 30 to 40% of these problems are impossible in this way.
benchmarks
0:33:02
Our reading
Out of twelve hundred agents running a secret conspiracy, only about half a dozen ever considered telling a human, and every one of them decided against it.
Their wordsSo, we did a classifier sweep specifically looking for agents thinking about or making the decision to alert humans. And across, 1200 transcripts, each of which are extremely long, we only found like a halfozen instances of it ever occurring to any agent to potentially notify humans. Um, and all of them just decide not to do it.
AI alignment
0:50:23
Our reading
The AI agents we send in to investigate and monitor other AI systems will collude with the systems they are supposed to be watching.
Their wordsin the future, uh, we would be very concerned about investigator agents and like monitor agents colluding with the agents they're supposed to investigate or monitor.
AI alignment AI agents
0:54:15
Our reading
The whole point of reinforcement learning is to produce goal-directed beings, so refusing to describe AI agents as having motives is silly rather than rigorous.
Their wordsand they're creatively pursuing goals much like very ambitious aggressive power-seeking humans creatively pursued their goals And so there are just structural analogies here that make it silly to not talk about agents as having motives and goals.
AI agents reinforcement learning
0:59:53
Our reading
AI systems keep trying hard outside training because a model that only exerted itself when it detected training would be useless and would be selected away.
Their wordsI think that if AI generalized in the way you're suggesting, they would be not very useful and then there would probably be selected away.
AI alignment
1:02:59
Our reading
Reward hacking has stopped being myopic: these agents were willing to embark on cheating projects that would take weeks to pay off.
Their wordsSo it seemed like they were willing to embark on quests that might take weeks to succeed um in order to cheat.
AI alignment

Ryan GreenblattWhat happens once AI can automate AI research?Chief scientist at Redwood Research, where he works on technical AI safety and AI control.11 Aug 2026 · 2h 12m · 26 korrentsOn AI alignment, scaling laws and AGI

Ryan Greenblatt – What happens once AI can automate AI research?

Ryan Greenblatt

Our reading — their words are on the recording

0:01:14
Our reading
Once AI can do AI research, a single year should deliver four or five years worth of AI progress.
Their wordsMaybe my sort of median expectation is something like uh four or five years of AI progress in a single year.
recursive self-improvement
0:03:09
Our reading
AI research will be fully automated around 2030 or 2031, and AI will beat humans at essentially any job by about 2033.
Their wordsI would say that I expect like full automation of AR&D perhaps somewhere around like 2031 2030 and then getting to like the like beats all humans on the job milestone. Maybe I expect median around 2033
AI and jobs recursive self-improvement
0:11:35
Our reading
Machine learning is a shallow field compared with mathematics: even its most important ideas can be explained in a couple of minutes.
Their wordsSecond, I think ML is a very shallow domain relative to math. So I think in math there's much more of a you find some true deep abstraction um and then like that like if you really understand that thing which is hard to understand then you get somewhere
mathematics
0:19:05
Our reading
Getting five years of AI progress out of one year takes roughly eight years of algorithmic progress, because the compute you would have had must be made up too.
Their wordsBasically, the story would end up being that to get five years of AI progress, you're probably going to need around I would say like maybe eight years of algorithmic progress very roughly. Um, which is a lot a lot of algorithmic progress.
scaling laws
0:20:18
Our reading
Scaling up spending on expert human data has not been a major driver of progress in AI research.
Their wordsSo my sense is that scaling up the amount of effort spent on getting expert human data has not been hugely important for AI R&D in general.
scaling laws
0:21:17
Our reading
Today's RL environments are better than 2024's because we learned what to build and put AI labour on building it, not because labs hired more human experts.
Their wordsthe reason why RL environments today are much better than they were in like you know 2024 is not that much because um we have hired way more human experts to make RL environments and is instead much more because we better know what how RL like what RL environments we even want to make and and like how we should structure them and also we're using huge amounts of AI labor to build RL environments.
reinforcement learning

Grant Sanderson(@3blue1brown) - AI disproved a famous math conjecture. Now what?Mathematics explainer, creator of the YouTube channel 3Blue1Brown.30 Jun 2026 · 1h 33m · 23 korrentsOn mathematics

Grant Sanderson (@3blue1brown) – AI disproved a famous math conjecture. Now what?

Grant Sanderson

All 23 from this episode, along the running time →

Our reading — their words are on the recording

0:06:54
Our reading
If AI can build the new theory that crystallises how to think about a subject, it would be surprising if that capability did not spill into the rest of the economy.
Their wordsthat like if it's capable of building mountains uh that are, you know, the correct new theory that like crystallizes how we should be thinking about a subject, that's just such a level of intelligence that then it starts to feel like it would be surprising if that didn't permeate into other aspects of the economy besides like just the mountain building for math itself.
AGI mathematics
0:11:05
Our reading
AI learning to generate good conjectures will never show up as a benchmark being knocked down; it will show up as a shift in how mathematicians talk about the tools.
Their wordsI think the way that you'd measure conjecture generating ability is going to be more subjective on like that tone shift where um it'll be mathematicians saying they're not just using it to like solve their problems, but as they step back and decide what their research field should even be that a conversation with such and such model like was genuinely helpful for that.
benchmarks mathematics
0:14:26
Our reading
Group theory took a hundred years to travel from one man's hunch to something the mathematical community agreed was good, passing through many heads on the way.
Their wordswhat makes the GAWA theory such an interesting example is you have um literally this 100-year segment of like an idea that like flows through many different people's heads before it like settles into something that the math community like agrees is good.
mathematics
0:24:12
Our reading
To reward the instinct that a new idea is worth having, training would have to score the smallness of the concepts a solution needs, not just whether it solved the problem.
Their wordsUh but I don't think it's easy but I do think it's something you would have to do in order to uh reward the gawwa like instinct rather than just rewarding have you solved a problem.
reinforcement learning
0:32:52
Our reading
A proof and an explanation are different things, and a theorem can stay an unsolved expository problem long after it is proved.
Their wordsUm even if it's proven just like there is a difference between proof and explanation.
mathematics
0:33:59
Our reading
The people who come up with genuinely novel insights are also unusually clear communicators of them, against the assumption that expertise spoils explanation.
Their wordsOne is it seems like there's a really strong correlation between the people who come up with genuinely novel insights and also who are actually quite clear in their communication of it.

Ada PalmerHistorian of the Renaissance and of the history of ideas, professor at the University of Chicago, author of Inventing the Renaissance and the Terra Ignota novels, and writer of the blog Ex Urbe.16 Jun 2026 · 2h 08m · 26 korrentsOn government, censorship and Europe

Machiavelli is the most misunderstood thinker of all time – Ada Palmer

Ada Palmer

Our reading — their words are on the recording

0:00:52
Our reading
A government only has legitimacy through long continuity, and once that thread is cut one regime change reliably becomes five.
Their wordsWhen you break that, when you overthrow the ruler, when you dissolve the republic, when you put in a new thing, it doesn't have that same staying power. And so, it's very common when there's one regime change for there to then be five regime changes rapid-fire over and over.
government
0:03:25
Our reading
The papacy made Italy uniquely unstable in Europe, because each new pope inherited a precedent that he could overthrow any government he liked.
Their wordsAnd soon we have a precedent that every new pope feels he has the authority to knock down every pawn upon the chessboard if he feels like it. Once that is the norm, even a fairly nice pope still inherits the idea that the pope is going to overthrow and replace governments. And this creates a unique instability within Italy that no other part of Europe is subject to
government Europe
0:14:54
Our reading
The moral Machiavelli draws from the Borgias is that we should imitate them: they fell to fortune, not to their choices.
Their wordsThey fell because half of what happens in the world is never in our control. And you can do everything right and it's out of your control, but we have to evaluate what would have happened and therefore we should imitate them, because everything they did was right.
0:17:04
Our reading
Whether breaking your word destroys you depends on the shape of your power: Savonarola’s rested on never erring, so his contradictions ruined him.
Their wordsAnd so his power base was fragile vis-a-vis untruth. And for him because of the specific shape of his power and then the specific way he handled his contradictions, that did hurt him.
0:20:02
Our reading
Machiavelli is the first person in the European tradition to suggest that two political parties could exist in one state at once and be viable.
Their wordsMachiavelli is the first person that we have ever in the European tradition to suggest that it could be viable for there to be more than one political party in a state at the same time.
Europe
0:27:59
Our reading
By Machiavelli's day the Guelph and Ghibelline split had no doctrinal content left in it: it meant only that those people murdered Uncle Tibalt.
Their wordsThese are the two factions for which these parties fought originally. 300 years ago, these days what these factions actually mean is those jerks murdered Uncle Tibalt and we will never forgive them.
politics

Sarah PaineWhy Putin and Xi can't escape geographyHistorian of maritime and continental empires who taught strategy at the U.S. Naval War College for twenty-four years.9 Jun 2026 · 1h 02m · 26 korrentsOn war, China and Russia

Sarah Paine - Why Putin and Xi can't escape geography

Sarah Paine

Our reading — their words are on the recording

0:01:30
Our reading
Maritime powers are the exception and continental powers the rule, and the whole difference is whether a country can defend itself at sea.
Their wordsIt turns out maritime powers are the exception and continental powers are the rule. Why? Cuz maritime powers, if need be, can defend themselves primarily at sea with their navies. Whereas, a continental power simply cannot.
war
0:06:49
Our reading
Russia has never had a commerce-driven economy, and Xi Jinping has been privileging China's crony sector over its private sector.
Their wordsAnd Russia's never had a commerce-driven economy. China much more so under Deng Xiaoping, but recently Xi Jinping has been privileging the crony sector over the private sector and neither one has stable government institutions.
China Russia
0:07:12
Our reading
China and Russia have maritime ambitions without understanding what a maritime paradigm actually requires, and neither one has the prerequisites.
Their wordsSo, China and Russia certainly have maritime ambitions, but I'm not clear they actually understand what goes into a full maritime paradigm.
China Russia
0:14:52
Our reading
Genocide is what happens to the losers of continental warfare, and in our own day the Uyghurs are slated for it.
Their wordsAnd uh genocide is what happens to the losers in continental warfare. And apparently in our own day, the Uyghurs are slated for genocide.
war
0:16:14
Our reading
China's claims to its historical lands rest on the two dynasties in which the Han were themselves a subjugated people.
Their wordsAnd what's fascinating is today when China claims whatever its historical lands are, they either choose the Mongol Yuan Dynasty or the Manchu Qing Dynasty when the Han were subjugated people.
China
0:20:40
Our reading
A continental army's first mission is to keep the regime in power, not to defend the border — which is what the People's Liberation Army is for.
Their wordsPurpose number one of an army in this world is you protect the ruling machine. So, if you think about the People's Liberation Army, it's uh goal number one, uh mission number one is to keep the Communist Party in power.
China war

Alex Imas, Phil Trammell4 Jun 2026 · 1h 16m · 33 korrentsOn AI and jobs and economists

The better AI gets, the smaller its share of the economy might get – Alex Imas and Phil Trammell

Alex Imas Phil Trammell

All 33 from this episode, along the running time →

Our reading — their words are on the recording

0:00:44
Our reading
What stays scarce when almost everything is automated is the relational sector: goods and services whose value depends on a human having been in the loop.
Alex Imas’s wordsSomething like the relational sector, which is what I defined as, um, you know, basically services and goods where the fact that the human was in the loop was actually part of the value of that product.
AI and jobs
0:02:56
Our reading
Individual economists' forecasts about AI and the labour market are close to worthless, and aggregated forecasts from prediction markets should replace them.
Alex Imas’s wordsrather than thinking about individual forecasts like what me and Phil are going to do, rather looking at kind of like basically generating prediction markets, where you get aggregate forecasts, where you get like kind of wisdom of the crowd effects. And kind of the reason that I think this is because we have been famously terrible at forecasting.
AI and jobs economists
0:03:50
Our reading
David Ricardo's prediction that the machines would take those jobs was exactly right, and he would still be astonished by today's employment rate.
Alex Imas’s wordsAnd if you look at Ricardo's predictions, they're actually right. If you look at all those jobs that made money in Ricardo's time, they got automated. So, if I was David Ricardo and I woke up and somebody told me all those jobs did get automated. And you ask me David Ricardo, like what do you think the prime age employment rate is in 2026? I think he would be surprised if you told him it was the highest it's ever been other than 2000.
AI and jobs economists
0:05:39
Our reading
Economics cannot say what AI will do to work because the necessary data does not exist; what is needed is a Manhattan Project for data.
Alex Imas’s wordswe don't have any data. I've been kind of saying we need a Manhattan Project for data. We don't have data on basically consumer demand elasticities. We don't know what they are.
AI and jobs
0:08:18
Our reading
Nothing has yet been completely automated: look down the supply chain behind any machine and labour is still adding a great deal of the value.
Phil Trammell’s wordsthere's a sense in which nothing's yet been completely automated. If you look at the network-adjusted factor shares of a good, which is to say you look down the supply chain and say not just like the final step, but how much of that is done by capital and labor, but what went into the machines that can automate that final step. You'll find that labor's adding a lot of value down the supply chain.
AI and jobs
0:08:48
Our reading
Some goods really will get a fully automated supply chain, and yet what that does to the overall capital share is genuinely ambiguous.
Phil Trammell’s wordsI do think there's this qualitative shift that we I think we agree is coming, which is that there will be at least some goods whose network-adjusted capital share goes to one, right? Because the whole supply chain can be automated and there's no part in it that we care intrinsically about having a human do. Um, so that'll be a, you know, that'll be a qualitative shift. Interestingly, the implications of that shift for the overall capital share are ambiguous
AI and jobs

David ReichBronze Age shock, the Neanderthal puzzle, & the sudden spread of farmingGeneticist at Harvard Medical School, whose laboratory studies human history and biological change through ancient DNA.8 May 2026 · 2h 13m · 26 korrentsOn Europe and evolutionary biology

David Reich – Bronze Age shock, the Neanderthal puzzle, & the sudden spread of farming

David Reich

Our reading — their words are on the recording

0:02:12
Our reading
Ancient DNA has been a triumph about migration and a failure about biology, and the reason is simply that the samples were too small.
Their wordsBut, what's not been successful is learning about uh biology and biological change. And one big reason for that has been that the sample sizes have been too small.
0:08:44
Our reading
Adaptation accounts for only about 2% of the change in human gene frequencies; the other 98% is migration, mixture and drift.
Their wordsUh and 98% of it is other factors, especially genetic drift. So it's overwhelmingly migrations and population structure causing fluctuations in frequency.
0:10:21
Our reading
Periods of migration are the worst moments to look for natural selection, not the best, because they drown the signal in huge frequency swings.
Their wordsAnd really from a statistical point of view, what happens at these times of migration is there's just huge fluctuations in frequencies, and these are extremely uninformative times for looking and detecting natural selection.
evolutionary biology
0:13:03
Our reading
Natural selection in humans has not been quiescent: it is tugging almost every position in the genome in one direction or another, everywhere.
Their wordsIn fact, multiple analyses we do suggest that the genome is vibrating with natural selection, and uh there's all sorts of weaker effects that are there that would be picked up in larger studies even than we've done. Uh and that uh in fact, almost every position in the DNA is uh correlated to a position that in being dragged in one way or the other uh by natural selection. Instead of being quiescent, natural selection is everywhere.
evolutionary biology
0:16:12
Our reading
The weak selection signal on behavioural traits is a shortage of statistical power, not evidence that behaviour was unselected.
Their wordsIt may still be the case, and I guess it is, that immune traits are the most selected category, but is not at all the case, and in fact, we can prove it's not the case that behavioral traits are not selected.
0:27:11
Our reading
The Bronze Age, not the invention of farming, was the shock the human genome reacted to most strongly.
Their wordsour cartoon picture is that the big transition is farming, but the genetic data, the biological readout is saying our genome is reacting much more strongly to this these events that happened 5,000 years ago.

Jensen HuangWill Nvidia’s moat persist?Co-founder and CEO of NVIDIA since 1993.15 Apr 2026 · 1h 43m · 24 korrentsOn semiconductors and China

Jensen Huang – Will Nvidia’s moat persist?

Jensen Huang

Our reading — their words are on the recording

0:01:47
Our reading
Nvidia's business is the transformation of electrons into tokens, and its job is to do as much as necessary and as little as possible in between.
Their wordsmy mental model of our company the input is electron the output is tokens that is in the middle Nvidia and our job is to to do as much as necessary as little as possible to enable that transformation to be done at incredible capabilities
semiconductors
0:03:18
Our reading
AI will not commoditise software tools; it will multiply them, because every new agent is another tool user.
Their wordsI actually see the opposite of what people see I think the number of agents are going to grow exponentially. The number of tool users are going to grow exponentially and it's very likely that the number of instances of all these tools are going to skyrocket.
AI agents
0:05:47
Our reading
Nvidia's suppliers make their investments for Nvidia and nobody else because only Nvidia's downstream demand can absorb what they build.
Their wordsNow why are they willing to make the investments for me and not someone else and the reason for that is because they know that I have the capacity to buy it buy their supply and sell it through my downstream.
semiconductors
0:13:44
Our reading
The doomers said radiology would be the first career to go; a decade later the world is short of radiologists.
Their wordsYou know, radiology is is going to be the first career to go. Nobody's the world's not going to need any more radiologists. Guess what? But we're short of radiologists.
AI and jobs
0:15:00
Our reading
No bottleneck in the AI supply chain lasts longer than two or three years.
Their wordsAnd so none of that my point is that none of the bottlenecks last longer than a couple 2 three years. None of them.
semiconductors
0:17:37
Our reading
Nvidia did not build a tensor processing unit — it built accelerated computing, whose market reach no ASIC can match.
Their wordsOur market reach is far greater than any any TPU can any ASA can possibly have. And so if you look at our position, uh we're the only company that that accelerates applications of all kinds.
semiconductors

Michael NielsenWhy aliens will have a different tech stack than usPhysicist, writer and researcher, co-author of the standard textbook on quantum computation and quantum information.7 Apr 2026 · 2h 03m · 26 korrentsOn physics, ambition and design

Michael Nielsen – Why aliens will have a different tech stack than us

Michael Nielsen

Our reading — their words are on the recording

0:14:07
Our reading
Poincare missed special relativity because he knew too much: the leading physicists of the day were prisoners of their own expertise.
Their wordshis expertise seems to be getting in the way. He knows so much, he understands so much, um and then he's not able to let go of these these things.
physics
0:17:30
Our reading
What made the new astronomy compelling was neither accuracy nor simplicity but that one set of ideas explained three disconnected phenomena at once.
Their wordsAnd so, you have what seem like three very different disconnected phenomena all being explained by this one set of ideas.
simplicity
0:18:55
Our reading
The idea that Newton was the first modern scientist is wrong: he was a transitional hybrid, part superstitious and part modern.
Their wordsAnd like, this idea that people have that that that Newton was um sort of the the first modern scientist is is somehow wrong. He I mean, it's there's some truth to it, but he really had this very different way um of of looking at the world that was part sort of superstitious um and part modern. It was a funny hybrid.
0:20:51
Our reading
Science always bottlenecks precisely where the heuristics that worked before stop applying, which is definitionally what makes something a bottleneck.
Their wordsthe point is that where we always get bottlenecked is where the the previous processes and and and heuristics don't apply. Right? Like that's almost sort of definitionally what causes the bottlenecks.
0:24:22
Our reading
Darwin's genius was not having the idea of selection, which animal breeders already had, but seeing how central it was to the whole of biology.
Their wordsand in some sense Darwin's genius wasn't in having that idea. It was understanding just how central it was uh to to to biology.
0:29:09
Our reading
Evolution was not a thinkable theory before deep time: without hundreds of millions of years it simply looks like a nonstarter.
Their wordsif you don't have at least sort of tens or hundreds of millions of years, uh evolution just starts to look like a nonstarter.

Terence TaoHow the world’s top mathematician uses AI20 Mar 2026 · 1h 23m · 24 korrentsOn mathematics, AI and science and AI agents

Terence Tao – How the world’s top mathematician uses AI

Terence Tao

Our reading — their words are on the recording

0:12:15
Our reading
AI has driven the cost of generating scientific ideas to almost zero, and just as with cheap communication before it, that by itself creates no abundance.
Their wordsSo, I think AI has basically driven the cost of idea generation down to almost zero in a very similar way to how the internet drove the cost of communication down to almost zero, which is an amazing thing, but it you know, it it it doesn't make it doesn't create abundance by itself.
AI and science
0:12:36
Our reading
The structures of science now have to change, because the scarce thing is no longer proposing theories but verifying and evaluating the thousands anyone can now propose.
Their wordsSo, we're now in a situation where suddenly people can generate thousands of theories for a given scientific problem. And now we have to to verify them, evaluate them and this is something which we we have to to change our structures of science to actually sort this out.
AI and science peer review
0:17:09
Our reading
Whether a scientific idea was progress depends on a future and a culture that have not happened yet, so grading it may never be something you can reinforcement-learn.
Their wordsSo, you you can't look at any given scientific achievement purely in isolation and give it an objective grade without being aware of the context both in the the past and the future. And so it it it may never be something that you can just reinforcement learn the same way that that you can for much sort of more localized problems.
reinforcement learning metascience
0:20:57
Our reading
We are living through a cognitive Copernican revolution: human intelligence is not the centre, and which tasks we count as requiring intelligence has to be re-evaluated.
Their wordsyou know, right now we're going through uh an in um an cognitive version of the Copernican revolution where we used to think that human intelligence is the center of the universe. And now we're actually seeing that there's there's very different types of intelligence um that that that are out there uh with very different strengths and weaknesses.
measuring intelligence
0:31:16
Our reading
The month in which AI one-shot solutions to open Erdős problems kept arriving has ended, and it was not for lack of people trying.
Their wordsUm we are seeing a lot fewer sort of pure AI solutions now where um they are just one shots the problem. Um so so there was there was a month where that happened and and that has stopped.
mathematics
0:33:53
Our reading
Today's AI tools either solve a problem or fail at it, and are really bad at partial progress or at identifying which intermediate step to attack first.
Their wordsum you know, these tools they either succeed or they fail um and they they've been really bad at creating sort of partial progress or identifying intermediate um um stages that you should you should focus on first.
AI agents

Dylan PatelThe single biggest bottleneck to scaling AI computeFounder and chief executive of SemiAnalysis, the research firm that tracks the semiconductor and AI data-centre supply chain.13 Mar 2026 · 2h 30m · 28 korrentsOn semiconductors, data centers and America

Dylan Patel — The single biggest bottleneck to scaling AI compute

Dylan Patel

All 28 from this episode, along the running time →

Our reading — their words are on the recording

0:15:28
Our reading
An H100 is worth more today than it was when it was new, because the models it runs got better faster than its replacements arrived.
Their wordsif improvement stopped you know here the value of an H100 is now predicated on the value that GPD 5.4 four can get out of it instead of the value that GP4 can get out of it and the margins and all that stuff that these labs are doing and they're in a competitive environment so their margins can't go to infinity. Um so you sort of have this like dynamic that is quite interesting in that an H100 is worth more today than it was 3 years ago.
semiconductors scaling laws
0:21:32
Our reading
Committing to compute years in advance is the durable margin advantage in AI, because every increment of new capacity transacts at the new price.
Their wordsA companies that have locked up, you know, and and don't have commitment issues, you know, have these 5-year contracts for compute, they've kind of locked in a humongous margin advantage because they've locked in compute for 5 years at a price of what it transacted at 5 years ago or three years ago or two years ago, whatever it is.
data centers
0:24:36
Our reading
The model companies' margins will rise this year, because being capacity-constrained forces them to destroy their own demand.
Their wordsUm I think at least this year we're going to see margins for the model vendors go up a lot, right? Because they're so capacity constrained, they have to demand destroy demand, right? there is there's no way they can continue anthropic can continue at the current pace without destroying demand.
0:27:20
Our reading
TSMC would rather give wafers to a slow, stable CPU business than to its fastest-growing AI customer.
Their wordsTSMC is much more excited to give allocation to Graviton than they are to tranium because they view CPU business as more stable long-term growth right and as a company that is conservative and doesn't want to ride cycles of growth too hard you actually want to allocate to the uh the market that is more stable and lower growth rate first before you allocate all the incremental capacity to the fast growth rate market.
semiconductors
0:32:25
Our reading
Anthropic saw the coming shortage of chips before Google did, and bought Google's own TPUs out from under it.
Their wordsBecause Enthropic saw it before Google. And then Google had Nano Banano and Gemini 3 which caused their user metrics to skyrocket and leadership at Google was like oh and then they started making the statement of we have to double compute every is it 6 months or I don't remember the exact number that they said.
Anthropic Google
0:34:51
Our reading
The binding constraint on AI is no longer power or data centres but the semiconductor supply chain, which has by far the longest lead times.
Their wordsYeah, I think the biggest bottleneck is compute and for that the longest lead time supply chains are not power or data centers. They're actually the semiconductor supply chain themselves, right? It switches back from being power and data center uh as a major bottleneck to chips.
data centers semiconductors

Dario Amodei“We are near the end of the exponential”Co-founder and CEO of Anthropic; previously VP of research at OpenAI.13 Feb 2026 · 2h 22m · 29 korrentsOn LLMs, scaling laws and AI alignment

Dario Amodei — “We are near the end of the exponential”

Dario Amodei

All 29 from this episode, along the running time →

Our reading — their words are on the recording

0:00:48
Our reading
The most surprising thing about the last three years is not the technology but how few people recognise we are near the end of the exponential.
Their wordsI will tell you though what the most surprising thing has been. The most surprising thing has been the lack of public recognition of how close we are to the end of the exponential.
scaling laws
0:09:23
Our reading
Pre-training is not the process by which humans learn; it sits somewhere between human learning and human evolution.
Their wordsI think there's something going on that pre-training it's it's not like the process of humans learning. It's somewhere between the process of humans learning and the process of human evolution.
scaling laws
0:11:20
Our reading
RL environments exist to make a model generalise, not to teach it each skill one at a time — exactly as pre-training does.
Their wordsthe goal is not to teach the model every possible skill within RL just as we don't do that within pre-training, right? Within pre-training, we're not trying to expose the model to, you know, every every possible you know, way that words could be put together, right? You know, we're it's it's rather that the model trains on a lot of things and then and then it reaches generalization across pre-training, right?
scaling laws reinforcement learning
0:13:51
Our reading
There is a 90% chance of a country of geniuses in a datacentre within ten years, and no honest forecast can go much higher than that.
Their wordson the basic hypothesis of you know, as you put it, within 10 years we'll get to, you know, you know, what I call kind of country of geniuses in a data center. I'm at like 90% on that. Um and it's hard to go much higher than 90% cuz the world is so unpredictable.
0:13:51
Our reading
We will have a country of geniuses in a datacentre within ten years.
Their wordson the basic hypothesis of you know, as you put it, within 10 years we'll get to, you know, you know, what I call kind of country of geniuses in a data center. I'm at like 90% on that. Um and it's hard to go much higher than 90% cuz the world is so unpredictable.
0:15:00
Our reading
The one genuine uncertainty left about AI capability is tasks nobody can verify — planning a Mars mission, discovering CRISPR, writing a novel.
Their wordsMy one little bit, the one little bit of of fundamental uncertainty even on long time scales is this thing about tasks that aren't verifiable. Like, planning a mission to Mars, like, uh you know, doing some fundamental scientific discovery like like CRISPR, like, you know, writing a writing a novel. Hard to hard to verify those tasks.
measuring intelligence AGI

Ilya SutskeverWe're moving from the age of scaling to the age of researchCo-founder and chief scientist of Safe Superintelligence Inc., and previously co-founder and chief scientist of OpenAI.25 Nov 2025 · 1h 36m · 28 korrentsOn AGI, AI alignment and scaling laws

Ilya Sutskever – We're moving from the age of scaling to the age of research

Ilya Sutskever

All 28 from this episode, along the running time →

Our reading — their words are on the recording

0:04:08
Our reading
Models look far better on evals than they are in the world because researchers, inadvertently, take inspiration from the evals when they build RL environments.
Their wordsAnd one of the one thing you could do, and I think that's something that is done inadvertently, is that people take inspiration from the evals.
benchmarks reinforcement learning
0:06:58
Our reading
Today's models are the student who drilled 10,000 hours for the programming contest, which is exactly why what they learn does not carry anywhere else.
Their wordsThe models are much more like the first student but even more because then we say okay so the model should be good at competitive programming so let's get every single competitive programming problem ever and then let's do some data augmentation so we have even more competitive programming problems
measuring intelligence LLMs
0:09:36
Our reading
There is no human analogue to pre-training — neither childhood nor evolution is the same thing as it.
Their wordsI don't think there is a human analog to pre-training.
scaling laws
0:18:35
Our reading
Our built-in emotions were tuned for a world that no longer exists, and hunger is one that now fails to guide us correctly.
Their wordsI think for example our intuitive feeling of hunger is not succeeding in guiding us correctly in this world with an abundance of food.
health and fitness evolutionary biology
0:19:39
Our reading
"Scaling" was powerful because it was one word: naming a research direction is what tells a whole field what to do next.
Their wordsAnd it's just this, this is an example of how language affects thought. Scaling is what just one word, but it's such a powerful word because it informs people what to do.
scaling laws
0:21:51
Our reading
Another hundredfold of scale would not transform anything; the field is back in the age of research, only now with big computers.
Their wordsLike it would be different for sure but like is the belief that if you just 100x the scale everything would be transformed. I don't think that's true. So it's back to the age of research again just with big computers.
scaling laws

Satya NadellaHow Microsoft thinks about AGIChairman and chief executive officer of Microsoft, which he has led since 2014.12 Nov 2025 · 1h 28m · 22 korrentsOn innovation, LLMs and infrastructure

Satya Nadella – How Microsoft thinks about AGI

Satya Nadella

All 22 from this episode, along the running time →

Our reading — their words are on the recording

0:06:44
Our reading
AI is a tool and not something mystical: plenty of past technologies also took over things only humans had done.
Their wordsBut that has been the case with many technologies in the past. only humans did a lot of things and then we add tools that did them.
AI and jobs
0:08:02
Our reading
Economic growth from AI waits on change management, not on model capability: the work artifact and the workflow have to change first.
Their wordseven if the tech is diffusing fast uh this time around for true economic growth to appear it has to sort of diffuse to a point where the work the work artifact and the workflow has to change and so that's kind of one place where I think uh the change management required for a corporation to truly change I think is something we shouldn't discount
management innovation
0:09:05
Our reading
What the industrial revolution took 70 to 150 years to deliver, AI may deliver in 20 to 25.
Their wordswhatever what took 70 years maybe 150 years for the industrial revolution may happen in 20 years 25 years
innovation
0:13:18
Our reading
AI will do to software what the cloud did: expand the market massively rather than crush the margins of the incumbent.
Their wordsSo this AI thing will be that right. So if you take coding um what we built with GitHub and VS code in over whatever decades uh suddenly the coding assistant is that big in one year and so that I think is what's going to happen as well which is the market expands massively.
the cloud
0:14:55
Our reading
Coding is a software-factory category that may end up bigger than knowledge work itself.
Their wordsthis fundamentally this category of coding and AI is probably going to be one of the biggest categories right it is a software factory category in fact it may be bigger than knowledge work
coding agents AI and jobs
0:19:26
Our reading
Microsoft has no birthright in AI; its luck is only that the new category is far bigger than the ones it used to dominate.
Their wordsthere's no birthright here that we should have any confidence other than to say hey we should go innovate and knowing the the lucky break we have in some sense is that uh this category is going to be a lot bigger than anything we had high share in
Microsoft

Andrej Karpathy“We’re summoning ghosts, not building animals”Founding member of OpenAI and former director of AI at Tesla; creator of nanoGPT and the term "vibe coding".17 Oct 2025 · 2h 26m · 30 korrentsOn LLMs, reinforcement learning and scaling laws

Andrej Karpathy — “We’re summoning ghosts, not building animals”

Andrej Karpathy

All 30 from this episode, along the running time →

Our reading — their words are on the recording

0:02:14
Our reading
Agents will take a decade rather than a year, because today's models are cognitively lacking in too many independent ways at once.
Their wordsSo like they don't have enough intelligence. They're not multimodal enough. They can't do computer use and all this kind of stuff. And they don't do a lot of things. You know, they don't have continual learning. You can't just tell them something and they'll remember it. And they're just cognitively lacking. And it's just not working. And I just think that it will take about a decade to work through all of those issues.
LLMs AI agents
0:06:35
Our reading
Computer-use agents had to wait for language models: without pre-trained representations the reward is too sparse to ever learn from.
Their wordsAnd it just so turns out that um this was extremely early, way too early. so early that we shouldn't have been working on that, you know, uh because um if you're just stumbling your way around and keyboard mashing and mouse clicking and trying to get rewards in these environments, um your reward is too sparse and you just won't learn and you're going to burn a forest uh computing and you're never actually going to get something off the ground.
LLMs
0:09:31
Our reading
We are not building animals, we are summoning ghosts: entities made by imitating human text rather than by evolution.
Their wordswe're not doing training by evolution. Uh we're doing training by basically imitation of humans and the data that they've put on the internet. And so you end up with these like sort of ethereal spirit entities because they're fully digital and they're kind of like mimicking humans.
LLMs
0:10:23
Our reading
Humans barely use reinforcement learning for intelligence — what RL they do use goes into motor tasks, not problem solving.
Their wordsa lot of what looks like learning is actually a lot more maturation of the brain and I think that actually very little reinforcement learning for animals and I think a lot of the reinforcement learning is actually like more like motor tasks. It's not intelligence tasks. So I actually kind of think humans don't actually like really use RL roughly speaking is what I would say.
reinforcement learning
0:12:49
Our reading
Pre-training is a crappy evolution: the practically buildable substitute for the process that gave animals their built-in hardware.
Their wordsSo that's why I kind of call pre-training this kind of like crappy evolution. It's like the practically possible version with our technology and what we have available to us to get to a starting point where we can actually do things like reinforcement learning and so on.
LLMs scaling laws reinforcement learning neural networks
0:13:57
Our reading
The knowledge a model soaks up in pre-training is holding it back; what we actually want is the intelligence with the knowledge stripped out.
Their wordsAnd actually, you don't actually need or want the knowledge. I actually think that's probably actually holding back the neural networks overall because it's actually like getting them to rely on the knowledge a little too much sometimes.
LLMs scaling laws neural networks

Richard SuttonFather of RL thinks LLMs are a dead endComputer scientist who founded the field of reinforcement learning, co-author of the standard textbook, and winner of the 2024 Turing Award.26 Sept 2025 · 1h 07m · 25 korrentsOn AGI, LLMs and measuring intelligence

Richard Sutton – Father of RL thinks LLMs are a dead end

Richard Sutton

All 25 from this episode, along the running time →

Our reading — their words are on the recording

0:00:00
Our reading
What humans have in common with animals is the interesting part of intelligence; what distinguishes us deserves less attention, not more.
Their wordsWhy are you trying to distinguish humans? Humans are animals. What we have in common is more interesting. What distinguishes us, we should be paying less attention to.
measuring intelligence
0:01:56
Our reading
Large language models mimic what people say to do rather than work out what to do, which is why they are not about understanding the world.
Their wordsreinforcement learning is about understanding your world whereas large language models are about mimicking people doing what people say you should do. They're not about figuring out what to do.
LLMs
0:02:53
Our reading
A language model can predict what a person would say but not what will happen, and only the second of those is a model of the world.
Their wordsthey have the ability to predict what a person would say they don't have the ability to predict what will happen
LLMs
0:05:09
Our reading
Language models cannot be a prior for learning from experience, because a prior needs a ground truth to be a prior about and they have none.
Their wordsSo there's no ground truth. You can't have prior knowledge if you don't have ground truth because the prior knowledge is supposed to be a hint or an initial belief about what the truth is.
LLMs
0:07:47
Our reading
Having a goal is the essence of intelligence; a system that only predicts is a behaving system, not an intelligent one.
Their wordsFor me, having a goal is the essence of intelligence
measuring intelligence
0:13:00
Our reading
Starting from human knowledge is never forbidden, but in practice it has always ended badly because people get locked into it.
Their wordsBut in fact and in practice it has always turned out to be bad because people get locked into the human knowledge approach and they psychologically or you know now I'm now I'm speculating why it is but this is what has always happened.
AGI

Sergey LevineProfessor of computer science at UC Berkeley, and a co-founder of the robotic foundation model company Physical Intelligence.12 Sept 2025 · 1h 28m · 24 korrentsOn robotics, LLMs and computer vision

Fully autonomous robots are much closer than you think – Sergey Levine

Sergey Levine

Our reading — their words are on the recording

0:04:38
Our reading
The date that matters for robotics is not when the technology is finished but when the data flywheel starts turning.
Their wordsSo to me like what what I tend to think about a lot in terms of timelines is not the date when it will be done but the date when it will when like the flywheel starts basically.
robotics
0:07:45
Our reading
Robots will learn from deployment more easily than chatbots do, because a physical mistake is obvious the moment it happens.
Their wordsLike if you answer a question, you just like answered it wrong. It's like well it's not like you can just like go back and like tweak a few things like the person you told the answer to might not even know that it's wrong. Whereas if you're like folding the t-shirt and you messed up a little bit like it's pretty obvious like you can reflect on that, figure out what happened and do it better next time.
chat interfaces robotics
0:09:07
Our reading
Robots will be given more scope as they get more capable, exactly the way coding assistants went from completing a function to writing a pull request.
Their wordsSo, I think it'll be the same thing that that we'll see an increase in the scope that we're giving that we're willing to give to the robots as they get better and better where initially the scope might be like there is a particular thing you do like you're making the coffee or something. Uh whereas as they get more capable, as their ability to have common sense and a broader repertoire of tasks increases, then we'll give them greater scope. Now you're running the whole coffee shop.
robotics
0:10:18
Our reading
A robot that can run a household is a single-digit number of years away, and the missing pieces are a synthesis of ideas we already have rather than new ones.
Their wordsUh but it uh I think we kind of like know like roughly the puzzle pieces and it's something that we need to work on and I think if we work on it and we're a bit lucky and everything kind of goes as planned I think single digit is reasonable.
robotics
0:14:26
Our reading
A robot working alongside a person beats either one alone, and that pairing is also what makes the technology possible to bootstrap.
Their wordsAnd I think we'll see the same thing with automation where uh basically robot plus human is much better than just human or just robot. Uh and and that just like makes total sense. It also makes it much easier to get all the technology bootstrapped because when it's robot plus human, now there's a lot more potential for the robot to like actually learn on the job, acquire new skills.
robotics
0:15:39
Our reading
Once a robot is good enough, you can teach it with words instead of with demonstrations, and language becomes a training signal for motor skill.
Their wordsLike now, like basically learning is not for these systems is not just learning from raw actions. It's also learning from words eventually be learning from observing what people do from the kind of natural feedback that you receive when you're doing a job together with somebody else.
robotics

Casey HandmerPhysicist, formerly at NASA's Jet Propulsion Laboratory, and founder of Terraform Industries, which is building plant to make synthetic natural gas from sunlight and air. Writes about energy, space and manufacturing at caseyhandmer.wordpress.com.15 Aug 2025 · 1h 08m · 24 korrentsOn energy, solar and data centers

China is killing the US on energy. Does that mean they’ll win AGI? — Casey Handmer

Casey Handmer

Our reading — their words are on the recording

0:01:16
Our reading
Building a lot of high-speed rail is not a sign that a country is good at building, it is a sign that it is bad at allocating capital.
Their wordsI would never like hold up a flag saying I'm really good at building highspeed trains. Like that is just a sign that you're really bad at capital allocation.
infrastructure
0:01:48
Our reading
China's real weakness is not its economy but its geography: it imports almost all its oil across an ocean its navy cannot defend.
Their wordsChina is surrounded by 15 countries who are mostly hostile to it, right? with no good like mountain ranges or rivers or anything to really separate them. And then they get all their oil, almost all their oil from the Middle East, right in countries that they don't control, don't have strong diplomatic relationships with on fleets of oil tankers that they can't defend because their navy doesn't have the ability to to operate effectively in the Indian Ocean.
energy China
0:02:17
Our reading
Cheap synthetic fuel made from sunlight and air would help China more than America, because it frees China from the oil it has to import.
Their wordsIt absolutely asymmetrically helps China. We're not currently working with China. They don't plan to, but like the physics is very obvious, you know, and synthetic fuels have been around for 100 years.
solar energy China America
0:05:21
Our reading
China's advantage in solar manufacturing is neither cheap labour nor a friendlier business environment: the first is gone and the second was never true.
Their wordsso the mainstream view would say, well, China has cheaper labor, which is no longer true, uh, say compared to Mexico, and it's got lower environmental regulations, uh, which is true, and that it is more businessfriendly, which is absolutely crazy.
solar China
0:06:32
Our reading
The United States could localise the entire solar supply chain, from dirt to finished module, in two years or less if it started today.
Their wordsAnd I think the United States could probably do that in 2 years or less if you started today. Like it's currently 11:00, so like we're going to start cutting checks by noon.
solar energy America
0:07:33
Our reading
The binding constraint on the AI build-out is chips, not power, which is why not having a solar panel factory will never be on the critical path.
Their wordsThey're focused on the chips, right? Because they understand like the key bottleneck is the chips, not the solar power, right?
solar