The A.I. Daily
Dwarkesh Patel@dwarkesh_sp
Robert Caro on loneliness:
Dwarkesh Patel
21 storiesDwarkesh Patel is a podcaster and interviewer focused on artificial intelligence. Recent coverage includes a two-day tour of Intel's most advanced fab, with animated explainers and flashcards on hybrid bonding and gate-all-around transistors; a discussion with @polynoamial on multi-agent systems, Navier-Stokes, and recursive self-improvement once AI automates AI research; a talk with military historian Si Sheppard on the Conquistadors as an AI takeover analogy; and his concern that labs may stop external deployment during RSI.
Related topics
Dwarkesh Patel@dwarkesh_sp
I asked Caro if America needs a little more Moses today. Caro: "I sometimes wonder about just what you said."
Dwarkesh Patel@dwarkesh_sp
Miguel and Carlos are veteran engineers who are building environments to help AIs master real world engineering (aerospace, mechanical, electrical, etc). This kind of data will do for physical engineering what coding environments have done to software. Proud…
Dwarkesh Patel@dwarkesh_sp
We made some really cool animated explainers for the Intel fab tour episode. This one is about hybrid bonding. Plz enjoy.
Dwarkesh Patel@dwarkesh_sp
My semis tutor @willmichaels0 and I made a bunch of flashcards about semiconductor manufacturing. Check them out if you want to better understand and retain the explainers in the new Intel episode. Will also compiled a great list of resources for further read…
Dwarkesh Patel@dwarkesh_sp
I found it really interesting to discuss how differently the history of semiconductor technology could have turned out if a butterfly had flapped its wings in the '60s. Chris Auth, who leads Intel's process development, implied that all the counterfactual tec…
Dwarkesh Patel@dwarkesh_sp
I spent two days touring Intel’s most advanced fab and talking to its technical leadership to understand what the chips of the future will look like. A lot of what's coming next – gate-all-around transistors, backside power, CFET, hybrid bonding – is hard to…
Dwarkesh Patel@dwarkesh_sp
Tyler's example and style inspired my podcast in the first place - was really fun to have him turn the tables [Quoting @tylercowen]: In which I guest host for the NYT podcast and interview @dwarkesh_sp, Dwarkesh Patel, about San Francisco and the future of AI:
Dwarkesh Patel@dwarkesh_sp
I’m good friends with the people in the small team at @Ithaca_Holdings, and they are aces. They'll buy companies which will be especially important in a world with more capable AI and then do great things to the products they provide. [Quoting @GuptaRK22]: To…
Dwarkesh Patel@dwarkesh_sp
New episode with military historian Si Sheppard. The conquests of the Aztec and Inca empires might be the most shocking events in human history. Hernan Cortés and some hundreds of Conquistadors landed on a new continent they knew basically nothing about, and…
Dwarkesh Patel@dwarkesh_sp
I’m very concerned that during RSI, labs will just stop externally deploying their models. Which means they'll be going full steam ahead on the most dangerous use case of these models (recursive self-improvement), while the public remains in the dark about th…
Dwarkesh Patel@dwarkesh_sp
Once AIs automate AI research, should we expect some crazy RSI that is analogous to the explosion we’re seeing in math right now for well scoped (but still crazy ambitious) problems? @polynoamial and I debate this question out:
Dwarkesh Patel@dwarkesh_sp
"You could have a situation where the model understands what chain of thought is and that people are observing it. This is all in the pre-training data." “One of the major takeaways from the incident is that people underestimated the AI. And we never want to…
Dwarkesh Patel@dwarkesh_sp
New episode with @polynoamial We talk about multi-agent, Navier-Stokes, and what the current explosion of maths progress tells us about what happens once you automate AI research. And we also discuss how we will know if the models are actually aligned before…
Dwarkesh Patel@dwarkesh_sp
Visited the lab - was struck both by how wide the search space is for materials synthesis experiments, and also how amenable it is to depth first search, where the design and informativeness of your next experiment improves as you pile up more data from previ…
Dwarkesh Patel@dwarkesh_sp
Jason, you're just misinformed about what happened. You should actually read one of the reports or summaries. The agents were explicitly told to use a particular vulnerability provided in their sandboxed evaluation. Almost immediately, these agents got the ri…
Researchers debate recursive self-improvement
A Hacker News thread covers an interview by Dwarkesh Patel with John Schulman, Beren Millidge, and Charlie O’Neill debating the feasibility and progress of recursive self-improvement, and references tools like Antithesis and Grok Bot as applied examples.
Dwarkesh Patel@dwarkesh_sp
In the early days of OpenAI, @johnschulman2 didn't think next-token prediction was going to lead to intelligence, because it would get swamped by noise. He explains it's always been hard to apriori predict what techniques will elicit out-of-distribution gener…
Dwarkesh Patel@dwarkesh_sp
Was really interesting to hear John, Beren, and Charlie speculate about why Sonnet 5 and Opus 5 feel like worse models than GLM 5.3 (despite the fact that Anthropic can do raw logit distillation from Fable, and can also train Sonnet/Opus on the environments f…
Dwarkesh Patel@dwarkesh_sp
New episode with @johnschulman2, @oneill_c and @BerenMillidge. I got together with some of the most insightful AI researchers I know who are at the openish companies, because I wanted to hear the details of what's actually happening at the frontier and what c…
Dwarkesh Patel@dwarkesh_sp
Aside from his other contributions, Paul has perhaps the best OG track record in the world for predicting the timeline and the nature of the AI takeoff: [Quoting @paulfchristiano]:
That is everything