Long-form interviews, talks and roundtables worth the time. Mostly the person who built the thing, talking.
Altman on why now is the moment to start something. You sent this one.
The foundational Anthropic interview, both Amodeis, from 2022. Worth it for how much of the current strategy was already stated out loud.
Netflix CPTO on hiring systems thinkers over specialists in the AI era.
Invest Like The Best is one of your regulars. The most substantive Altman sit-down of the window.
Altman on company building rather than models.
David Senra doing the Founders treatment on Altman. You save Senra.
Bloomberg's extended Circuit cut. The long Dario interview you asked for.
The documentary version, 2.8M views. Access inside the building.
Stanford GSB. Hassabis is the most underrepresented of the three in most feeds.
Ryan Greenblatt on Dwarkesh, 2h12. The exact question Tworek answers with 'two years'. You save Dwarkesh and I had zero of him on the list.
Rich Sutton, who wrote the bitter lesson and won the Turing award for RL, on why models stop learning and how to restart it.
John Jumper. Nobel for AlphaFold, then walked. MLST does the most technical interviews going.
Ilia Shumailov on why labs hide reasoning traces. Pairs with the Apollo episode above.
Jerry Tworek, ex-OpenAI VP of Research, built o1/o3/Codex. Says human researchers have roughly two years before the loop closes. Best primary source of the month.
Karpathy's Stanford lecture. Rare full-length teaching, not an interview.
LeCun's case that world models, not LLM scaling, are the next step.
Noam Brown on reasoning and proof. He built the poker and Diplomacy systems and then o-series reasoning.
OpenAI's research chief on AGI, o1, evals and scaling laws. Latent Space is in your saves four times.
Anandkumar, Caltech. Already in your Watch Later, so you found this one yourself.
xAI co-founder. Already saved by you.
Bronson Shane of Apollo Research has read more frontier chain-of-thought than almost anyone. Finds constraint violations rising while labs report alignment improving.
DeepMind's own interpretability explainer.
Sequoia. Short, and continual learning is the actual bottleneck for agents that persist.
Toby Ord, Oxford. Nearly three hours on why timeline forecasts break.
80,000 Hours retrospective on a year of forecast revisions.
Jensen and Zuckerberg in the same room on what the next computing platform is. You saved it; it belongs in the queue.
The long Lex sit-down with Jensen, 2h26. The definitive one.
Altman with Lex on GPT-5, Sora and the board saga. The most candid he's been.
Kantrowitz is your 4th most-saved channel. Dario on the OpenAI rivalry and the business.
Dario on Dwarkesh, 2h. The most technical Dario interview that exists.
Hassabis on the AGI timeline, compact at 34 minutes.
Dylan Field on craft as the moat once everyone can generate. The counter-argument to commoditisation.
How Cursor got to $300M ARR, from the people who built it.
The long Armstrong interview, 2h29 with Lex.
Amjad Masad arguing the SaaS apocalypse is real. Read against Dario saying the opposite.
Stripe's series, looser format, Dario off the usual script.
Musk on Dwarkesh arguing compute goes orbital. The tail scenario Gavin Baker also raises.
Matthew Prince on the web's business model collapsing under agents. Directly upstream of agentic payments.
The operator's view from inside OpenAI rather than the founder's.
Logan Kilpatrick, DeepMind. Direct counter to the Harrison Chase harness talk. Read them together.
OpenRouter's CEO sees the actual routing demand across every model. Best available data on open vs closed share.
Lip Bu Tan on No Priors. Intel's CEO on rebuilding the supply chain.
Dylan Patel of SemiAnalysis on where the real 100x comes from.
Inference economics, which is the same thesis as the Sail Research episode from the supply side.
Anjney Midha. Compute is not the constraint people think it is.
Factory CTO Eno Reyes. Smartest model becomes the cheapest; 99% of workflows on open models in three years; calls the Chinese-models framing a lab psyop.
Vercel CEO Guillermo Rauch on the agent every company ships next.
Rauch again, on how Vercel itself operates. Pair with the one above.
Matei Zaharia. Spark creator, now betting Databricks on the agent cloud.
Eiso Kant of Poolside, the company NVIDIA just paid $6B to license.
Shipping velocity inside xAI, told by the people who did it.
Sail Research CEO, $80M from Kleiner and Sequoia. Token prices fall while enterprise bills triple because the stack was built for chatbots, not week-long agents.
Harrison Chase, LangChain. Directly useful for how you run Claude Code.
Kleiner Perkins. How Anthropic actually organizes its platform org. Already in your saves.
Hugging Face CEO on the shift from renting to owning intelligence.
Akshay Nathan, OpenAI. Where ChatGPT is heading as a surface.
Amjad Masad on building Replit to $9B.
Glean's founder making the platform-risk argument from the app side.
A second full DHH interview I completely missed, 1h41 with Senra on staying independent. You asked for DHH and this is the business half of it.
The one you're already watching. Five hours, and the reversal is genuine.
The hands-on technical companion to the Lex episode. Omarchy is the artifact behind his AI claims.
Addy Osmani. Pragmatic Engineer does the deepest engineering interviews going.
Charity Majors, Honeycomb CTO. The observability angle on agent-written code.
Cal Newport takes the rogue-agent story apart: it's a human-written harness looping prompts, not agency. Direct counterweight to Tworek.
Patrick Collison at YC on whether the playbook still holds.
a16z crypto is your single most-saved channel and I had zero of it. Andreessen and Dixon on the CLARITY Act.
Eddy Lazzarin, a16z crypto CTO. The structural stablecoin argument.
Morpho's CEO on why $200 trillion of credit moves onchain. Onchain lending from the protocol side, not a commentator's.
The sharpest argument against your agentic-payments thesis. Worth 18 minutes precisely because you disagree.
The longer Hayes piece, the cross-asset version.
Chainlink Labs. You save Chainlink and I had none.
Pollak on the app pivot, separate from the Rollup interview.
Carlos Domingo's structural claim about who gets to tokenize. Direct to the Coinbase-on-Base launch.
Bankless on the remaining institutional blockers.
Dan Boneh, Stanford crypto. The long-horizon risk nobody prices.
Searching for Mana is your single most-saved channel after a16z crypto. Circle's president on winter to IPO.
Base lead. Directly relevant with Coinbase launching tokenized equities on Base.
Stani Kulechov on Aave V4's design. The clearest statement of how onchain lending gets structured like a central bank.
The thesis you're building against, argued by Voorhees.
The longest recent Armstrong sit-down, 2h19, on bitcoin, banks and monetary rules.
Catalini, Libra co-creator. Consumer agentic commerce goes to Stripe; M2M is where crypto infra matters. Names x402.
Robert Leshner, Compound founder, on permissioning at the asset level instead of forking permissioned DeFi.
Securitize CEO Carlos Domingo: 2% of equities onchain is $2 trillion, the size of all crypto today.
Talking Tokens, another of your saves, on the same 2% framing.
Aerodrome CEO on Coinbase's tokenized equity launch on Base. Took 25% of AMM volume across all chains within days.
Sequoia. Jensen on building NVIDIA, 360K views. You asked for founders and I had none of him.
The YC version, different interview, more on mindset than strategy.
Kalanick on a16z about operating in stealth, 197K.
YC's president on what changed for founders.
Senra at Sequoia on the pattern across 400 founder biographies. You save Senra twice over.
Scale AI founder at YC.
Bg2 with Gavin Baker. Bg2 returned nothing in my search pool at all, which was a straight coverage hole.
Dylan Patel on Dwarkesh, 226K. Compute centralization, which is the supply question underneath every AI position.
Eric Vishria on Invest Like The Best, one of your regulars.
Martin Casado tracing where the AI money actually landed.
Jensen on All-In. Physical AI, the agent, and the inference explosion, from the person selling the shovels.
Four operators in one room: CoreWeave, Perplexity, Mistral, IREN. Compute, models and power in a single conversation.
Charles and Chase Koch on building to $150B while staying private. The anti-IPO case, from the largest example of it.
Cerebras and Black Forest CEOs on the open-source shift. Silicon and image models in one sitting.
Four investors pitching live trades. Pure idea generation, which is what you use the show for.
Loeb on short selling and why stock picking came back. Rare long-form from him.
Concrete enterprise AI result with a number attached, not a projection.
The IPO and valuation debate at length. Useful for the equity side.
Where the moats actually are, argued by four investors who are positioned.
The liquidity panel on why the IPO window reopened.
Dario in 2016, at Google Brain, years before Anthropic. The earliest substantial talk of his that exists on YouTube.
Dario at Microsoft Research in 2018, while he was at OpenAI. Pre-Anthropic thinking on safety and scaling.
Ilya on Lex in 2018 on meta-learning and self-play. Reads very differently now that self-play is central again.
OpenAI's 2019 deep-RL workshop, 3h25. The teaching material the current reasoning models grew out of.
Chris Olah in 2012, a TEDxYouth talk. He went on to co-found Anthropic and lead interpretability. The oldest thing on this list by four years.