Dwarkesh Podcast

Eric Jang walks through how to build AlphaGo from scratch, but with modern AI tools.
Sometimes you understand the future better by stepping backward. AlphaGo is still the cleanest worked example of the primitives of intelligence: search, learning from experience, and self-play. You have to go back to 2017 to get insight into how the more general AIs of the future might learn.
Once he explained how AlphaGo works, it gave us the context to have a discussion about how RL works in LLMs and how it could work better – naive policy gradient RL has to figure out which of the 100k+ tokens in your trajectory actually got you the right answer, while AlphaGo’s MCTS suggests a strictly better action every single move, giving you a training target that sidesteps the credit assignment problem. The way humans learn is surely closer to the second.
Eric also kickstarted an Autoresearch loop on his project. And it was very interesting to discuss which parts of AI research LLMs can already automate pretty well (implementing and running experiments, optimizing hyperparameters) and which they still struggle with (choosing the right question to investigate next, escaping research dead ends). Informative to all the recent discussion about when we should expect an intelligence explosion, and what it would look like from the inside.
Watch on YouTube. Read the transcript.
And check out the flashcards I wrote to retain the insights.
Sponsors
* Cursor‘s agent SDK let me build a pipeline to generate flashcards for this episode. For each card, I had an agent read the transcript, ingest blackboard screenshots, generate an SVG visual, and run everything through a critic. A durable agent is much better at this kind of work than a chain of LLM calls, and Cursor’s SDK made it easy. Check out the cards at flashcards.dwarkesh.com and get started with the SDK at cursor.com/dwarkesh
* Jane Street gave me a real deep-dive tour of one of their datacenters. I got to ask a bunch of questions to Ron Minsky, who co-leads Jane Street’s tech group, and Dan Pontecorvo, who runs Jane Street’s physical engineering team. They were willing to literally pull up the floorboards and take out racks to explain how everything works. Check out the full tour at janestreet.com/dwarkesh
Timestamps
(00:00:00) – Basics of Go
(00:08:17) – Monte Carlo Tree Search
(00:32:04) – What the neural network does
(01:00:33) – Self-play
(01:25:38) – Alternative RL approaches
(01:45:47) – Why doesn't MCTS work for LLMs
(02:01:09) – Off-policy training
(02:12:02) – RL is even more information inefficient than you thought
(02:22:16) – Automated AI researchers


Get full access to Dwarkesh Podcast at www.dwarkesh.com/subscribe

David Reich is back.
He and collaborator Ali Akbari just published a paper that overturns a long-standing consensus about human evolution — that natural selection has been dormant in our species since the agricultural revolution.
By scaling ancient DNA sequencing and developing a new statistical method, they found that selection has actually sped up.
Selection went especially bonkers during the Bronze Age (around 3,000 years ago).
That’s when gene frequencies for everything from immune function to body fat to intelligence were most in flux.
Over the last 10,000 years, selection pushed the genetic predictor of cognitive performance up by roughly a full standard deviation — most of it between 4,000 and 2,000 years ago.
After we finished recording, David sketched out on a whiteboard his new heretical model about who the Neanderthals really were. Luckily, I took out my iPhone and managed to record it.
He thinks the standard story (that Neanderthals are some separate archaic lineage we interbred with a little) just doesn’t fit the evidence. Instead, he proposes that Neanderthals are essentially genetically-swamped modern humans.
A small population somewhere around the Caucasus invented Middle Stone Age technology roughly 300,000 years ago and expanded outward. The ones that moved into Europe interbred with local archaic humans, got genetically swamped, and became Neanderthals. The same expansion went into Africa, met much more diverged archaic Africans, and that mixture became us.
This means Neanderthals and modern humans share the same cultural ancestry — the only difference is which archaic humans they mixed with afterward.
David is a brilliant and rigorous scholar. It was a real delight to learn from him again.
Watch on YouTube; read the transcript.
Sponsors
* Cursor was super useful as I prepped for this episode. Whenever I had a question, I’d have Cursor kick off a few different models simultaneously and then compare their responses. I found that this led to better results than I could get out of any individual LLM. If you’ve only used Cursor for coding, you should try using it for research. Check it out at cursor.com/dwarkesh
* Jane Street uses an internal currency called “hive bucks” to allocate compute through a real-time auction – and anyone can change anyone else’s bids or even kill their jobs! Everyone just trusts each other to act in the firm’s best interest, which is what lets the system work in the first place. If this weird and high-trust culture sounds like your kind of thing, Jane Street’s hiring at janestreet.com/dwarkesh
* Crusoe’s ML infra team built fastokens, an open-source tokenizer that delivers a ~9x speedup over Hugging Face and up to 40% faster time-to-first token – on real production workloads! Crusoe achieved these results by parallelizing things and using some clever engineering to handle duplicates without cross-thread coordination. Learn more at crusoe.ai/dwarkesh
Timestamps
(00:00:00) – Ancient DNA suggests strong selection over last 10,000 years
(00:15:45) – Natural selection intensified during the Bronze Age
(00:35:02) – Why didn’t evolution max out intelligence?
(00:57:21) – Evolution is limited by time, not population size
(01:09:02) – Why no farming before the Ice Age?
(01:17:13) – The Neanderthal puzzle David can’t stop thinking about
(01:54:10) – The methodology behind this breakthrough


Get full access to Dwarkesh Podcast at www.dwarkesh.com/subscribe

Did a very different format with Reiner Pope - a blackboard lecture where he walks through how frontier LLMs are trained and served.
It’s shocking how much you can deduce about what the labs are doing from a handful of equations, public API prices, and some chalk.
It’s a bit technical, but I encourage you to hang in there – it’s really worth it.
There are less than a handful of people who understand the full stack of AI, from chip design to model architecture, as well as Reiner. It was a real delight to learn from him.
Recommend watching this one on YouTube so you can see the chalkboard.
Reiner is CEO of MatX, a new chip startup (full disclosure - I’m an angel investor). He was previously at Google, where he worked on software efficiency, compilers, and TPU architecture.
Download markdown of transcript here to chat with an LLM.
Wrote up some flashcards and practice problems to help myself retain what Reiner taught. Hope it's helpful to you too!
Sponsors
* Jane Street needs constant access to incredibly low-latency compute. I recently asked one of their engineers, Clark, to talk me through how they meet these demands. Our conversation—which touched on everything from FPGAs to liquid cooling—was extremely helpful as I prepped to interview Reiner. You can watch the full discussion and explore Jane Street’s open roles at janestreet.com/dwarkesh
* Google’s Gemma 4 is the first open model that’s let me shut off the internet and create a fully disconnected “focus machine”. This is because Gemma is small enough to run on my laptop, but powerful enough to actually be useful. So, to prep for this interview, I downloaded Reiner’s scaling book, disconnected from wifi, and used Gemma to help me break down the material. Check it out at goo.gle/Gemma4
* Cursor helped me turn some notes I took on how gradients flow during large-scale pretraining into a great animation. At first, I wasn’t sure the best way to visualize the concept, but Cursor’s Composer 2 Fast model let me iterate on different ideas almost instantaneously. You can check out the animation in my recent blog post. And if you have something to visualize yourself, go to cursor.com/dwarkesh
Timestamps
(00:00:00) – How batch size affects token cost and speed
(00:32:09) – How MoE models are laid out across GPU racks
(00:47:12) – How pipeline parallelism spreads model layers across racks
(01:03:37) – Why Ilya said, “As we now know, pipelining is not wise.”
(01:18:59) – Because of RL, models may be 100x over-trained beyond Chinchilla-optimal
(01:33:02) – Deducing long context memory costs from API pricing
(02:04:02) – Convergent evolution between neural nets and cryptography


Get full access to Dwarkesh Podcast at www.dwarkesh.com/subscribe

I asked Jensen about TPU competition, Nvidia’s lock on the ever more bottlenecked supply chain needed to make advanced chips, whether we should be selling AI chips to China, why Nvidia doesn’t just become a hyperscaler, how it makes its investments, and much more. Enjoy!
Watch on YouTube; read the transcript.
Sponsors
* Crusoe’s cloud runs on state-of-the-art Blackwell GPUs, with Vera Rubin deployment scheduled for later this year. But hardware is only part of the story—for inference, Crusoe’s MemoryAlloy tech implements a cluster-wide KV cache, delivering up to 10x faster TTFT and 5x better throughput than vLLM. Learn more at crusoe.ai/dwarkesh
* Cursor helped me build an AI co-researcher over the course of a weekend. Now I have an AI agent that I can collaborate with in Google Docs via inline comment threads! And while other agentic coding tools feel like a total black-box, Cursor let me stay on top of the full implementation. You can try my co-researcher out at github.com/dwarkeshsp/ai_coworker, or get started on your own Cursor project today at cursor.com/dwarkesh
* Jane Street spent ~20,000 GPU hours training backdoors into 3 different language models, then challenged my audience to find the triggers. They received some clever solutions—like comparing the base and fine-tuned versions and extrapolating any differences to reveal the hidden backdoor—but no one was able to solve all 3. So if open problems like this excite you, Jane Street is hiring. Learn more at janestreet.com/dwarkesh
Timestamps
(00:00:00) – Is Nvidia’s biggest moat its grip on scarce supply chains?
(00:16:25) – Will TPUs break Nvidia’s hold on AI compute?
(00:41:06) – Why doesn’t Nvidia become a hyperscaler?
(00:57:36) – Should we be selling AI chips to China?
(01:35:06) – Why doesn’t Nvidia make multiple different chip architectures?


Get full access to Dwarkesh Podcast at www.dwarkesh.com/subscribe

Really enjoyed chatting with Michael Nielsen about how we recognize scientific progress.
It's especially relevant for closing the RL verification loop for scientific discovery.
But it's also a surprisingly mysterious and elusive question when you look at the history of human science.
We approach this question stories like Einstein (who claimed that he hadn't even heard of the famous Michelson-Morley experiment, which is supposed to have motivated special relativity, until after he had come up with the theory), Darwin (why did it take till 1859 to lay out an idea whose essence every farmer since antiquity must have observed?), Prout (how do you recognize that isotopes exist if you cannot chemically separate them?), and many others.
The verification loop on scientific ideas is often extremely long and weirdly hostile. Ancient Athenians dismissed Aristarchus's heliocentrism in the 3rd century BC because it would imply that the stars should shift in the sky as the Earth orbits the sun. The first successful measurement of stellar parallax was in 1838. That's a 2,000-year verification loop.
But clearly human science is able to make progress faster than raw experimental falsification/verification would imply, and in cases where experiments are very ambiguous. How?
Michael has some very deep and provocative hypotheses about the nature of progress. One I found especially thought-provoking is that aliens will likely have a VERY different science + tech stack than us. Which contradicts the common sense picture of a linear tech tree that I was assuming. And has some interesting implications about how future civilizations might trade and cooperate with each other.
Watch on Youtube; read the transcript.
Sponsors
* Labelbox researchers built a new safety benchmark. Why? Well, current safety benchmarks claim that attacks on top models are successful only a few percent of the time, but the prompts in those benchmarks don’t reflect how real bad actors actually write. You can read Labelbox’s research here. If this could be useful for your work, reach out at labelbox.com/dwarkesh
* Mercury has an MCP that lets you give an LLM access to your full transaction history, including things like attached receipts and internal notes. I just used it to categorize my 2025 transactions, and it worked shockingly well. Modern functionality like this is exactly why I use Mercury. Learn more at mercury.com
* Jane Street’s ML engineers presented some of their GPU optimization workflows at GTC, showing how they use CUDA graphs, streams, and custom kernels to shave real time off their training runs. You can watch the full talk here. And they open-sourced all the relevant code here. If this kind of stuff excites you, Jane Street is hiring — learn more at janestreet.com/dwarkesh
Timestamps
(00:00:00) – How scientific progress outpaces its verification loops
(00:17:51) – Newton was the last of the magicians
(00:23:26) – Why wasn’t natural selection obvious much earlier?
(00:29:52) – Could gradient descent have discovered general relativity?
(00:50:54) – Why aliens will have a different tech stack than us
(01:15:26) – Are there infinitely many deep scientific principles left to discover?
(01:26:25) – What drew Michael to quantum computing so early?
(01:35:29) – Does science need a new way to assign credit?
(01:43:57) – Prolificness versus depth
(01:49:17) – What it takes to actually internalize what you learn


Get full access to Dwarkesh Podcast at www.dwarkesh.com/subscribe