Watch / Listen / Take notes
Selections for podcasts and videos. ← Back
Podcasts
The AI Edge · WATCH · Read · Make
WATCH
The right inputs shape how you think. These are the channels and conversations I keep returning to — follow the people, then study the standout pieces.
Subscribe to these
A mix of product, research, and frontier-AI voices. Subscribe to a few and let the feed do the curating.
Lenny's Podcast
Lenny Rachitsky interviews world-class product leaders and growth experts for concrete, actionable advice on building, launching, and growing a product. The go-to show if you care about how things actually get shipped.
Lenny's Podcast on YouTubeMatthew Berman
Near-daily AI news, model breakdowns, and tutorials, distilled into clear 10–15 minute videos. The best single channel for staying current without spending an hour a night doom-scrolling releases.
Matthew Berman on YouTubeLex Fridman
Long-form, unhurried conversations with researchers, founders, and thinkers across AI, science, and beyond. The marathon interviews give ideas room to breathe — great for deep background, not quick takes.
Lex Fridman on YouTubeDwarkesh Patel
Unusually well-researched interviews with leading AI researchers and historians. Dwarkesh asks the sharp, specific questions most interviewers miss — you come away genuinely understanding the guest's worldview.
Dwarkesh Patel on YouTubeThe Cognitive Revolution
Nathan Labenz tracks how AI is reshaping the world through interviews with builders and researchers. Practical and timely, with a builder's eye for what's actually shipping and what it means.
The Cognitive Revolution on YouTubeTBPN
A fast-moving, live daily show covering tech and AI business news with a markets-and-startups lens. Good for the industry pulse — who raised, who shipped, what the Valley is talking about today.
TBPN on YouTubeSpecific pieces worth your time
Individual talks and interviews I'd assign — sit down, take notes, and watch with intent.
Ilya Sutskever on the limits of scaling
Part 1 looks at the effect of reinforcement learning on models — they perform well on evals, yet the real-world economic impact stays limited. Part 2 argues we're moving from the age of scaling into an age of research, where new ideas matter more than raw compute. A clear-eyed take from one of the field's founding figures.
Watch the interviewAleph and energy-based models: the AI that refuses to bullshit
An accessible look at energy-based models — an alternative to the next-token-prediction paradigm behind today's LLMs — and why that architecture is structurally less prone to confidently making things up. A good companion to the JEPA piece below.
Watch the talkReproducing LeCun's JEPA world model that doesn't predict tokens
A from-scratch reproduction of Yann LeCun's recent world model — roughly 15M parameters, trainable in a few hours, with no foundation-model dependency. Watch this to see the JEPA idea (predict in representation space, not token space) become real, runnable code.
Watch the reproductionThe hardware lottery
Hooker's influential argument that research ideas often win not because they're best, but because they happen to fit the hardware and software of the moment. Essential context for understanding why the field looks the way it does — and connects directly to Step 7's local-LLM hardware lessons.
Watch the talkThe notebook habit
Get a paper notebook. Not digital — the friction matters. After every episode or video, log:
One big idea worth keeping
Three unknown terms or concepts you need to look up
Then look them up. That habit alone separates people who learn from people who consume.
