Free — sign in with your email, no password.
Podcasts on 2 October 2026
AI-native startups can beat incumbents by owning the full end-to-end job, not just a system of record, and procurement is a prime place to do it.
12 key points · 1 quoteAI discourse is polarized and gloomy, but a growing group of moderates, including Fukuyama, Kapoor and Narayanan, and Katzenberg, reject false binaries and focus on specific problems and terms.
8 key points · 2 quotesSign-in with ChatGPT hints at a new bring-your-own-compute model, while the hosts and guests discuss AI and religion, Tesla self-driving, a unified government website, campus-scouted VC, and Larry Ellison's strategy.
10 key points · 1 quoteNvidia's plan to aggregate neoclouds fizzled into a friendly partnership, while guests from Varda, Ecolab and Tavus argue that fleets, closed-loop cooling and human-like AI models are becoming real products.
10 key pointsMike and Liv argue that AI-safety politics have been polarized by tribal fights and strange alliances, that data-center backlash is really about digital life, and that media and feminist narratives dodge personal responsibility.
7 key points · 1 quotePersonal AI agents are arriving just as rogue agents, sloppy lab security and a scrapped OpenAI model expose how weak self-regulation is, while consumers trade privacy for convenience.
9 key points · 1 quoteEight entrepreneurs compete for $100,000 and a year of mentorship. The episode argues that conviction, fast execution of a clear plan, and sound sales judgment can produce years of growth in 90 days.
7 key points · 1 quoteJonathan argues AI training has moved from teaching tests to mastering real work via simulated environments, that enterprises should own their learning loops with open-weight models, and that takeoff will be slow.
9 key pointsGPU-and-backprop computing is plateauing on energy efficiency, and talks on zero-order sharded training, optical compute, neuromorphics and living neurons explore substrates co-designed with their algorithms.
9 key pointsWalter Goodwin argues frontier AI inference needs chips with very high memory bandwidth and affordable capacity, built by lean full-stack teams that iterate fast, since frontier labs cannot safely bet alone on proprietary silicon.
8 key points · 2 quotesClusterMax 3.0 rates GPU clouds on managed-cluster quality, finds neocloud security and compute availability bad, and shows labs, Nvidia and chip makers reshaping who supplies and buys compute.
10 key points · 4 quotesAlex Zhang argues that harnesses are the underexplored lever: recursive language models, which offload context into code, generalize better, and academics should take big bets on simple ideas that look trivial.
8 key pointsThe hosts argue that the White House accord makes AI safety governance practical without pausing development, that compute and power are national security assets, and that doomerism and media narratives distort the economy and politics.
10 key points · 2 quotes