Channels | LLM Bento

Lex Fridman

Three hours is not a format, it's a filter. What Fridman does that shorter shows can't is wait — through the hedge, past the prepared answer, until the guest says something they hadn't planned to. The range is the other reason this channel outweighs every other source here by a factor of two: Karpathy and Hassabis sit in the same catalogue as Kahneman, Dawkins, and Vervaeke, which is how you notice the AI conversation was never only an AI conversation.

30 conversations here

Dwarkesh Patel

@DwarkeshPatel

Patel prepares like someone who expects to be caught out — reading the papers, the earlier interviews, the post from four years ago the guest has forgotten writing, then asking the question that assumes you've read them too. What this collection takes from Patel skews toward agents and alignment because that is where the attention has been: Sutskever on what comes after scaling, Sutton arguing LLMs are a dead end, Aschenbrenner on a 2027 intelligence race, Kokotajlo and Scott Alexander modelling the explosion month by month.

13 conversations here

80,000 Hours

@eightythousandhours

A nonprofit that researches which careers actually help, running a podcast to work out what the answer is — which turns out to be the most useful constraint a long-form AI show can have. Nothing here is allowed to stay abstract, because the question underneath every episode is what a listener should do on Monday. The four conversations this collection draws from are the whole cast of the pacing debate at length: Greenblatt laying out four takeover scenarios over nearly three hours, Cotra on what 10,000 years of progress compressed into decades would feel like, Kokotajlo on a hyperspeed robot economy, Bengio on whether safe superintelligence is even the right thing to aim at.

4 conversations here

Big Think

@bigthink

Big Think publishes both the clip and the whole conversation, and what this collection keeps is almost always the whole one — Brian Cox on the quantum roots of reality, Sean Carroll on the biggest ideas in the universe, an hour and a half on what the body knows. Most of it isn't about AI at all: Feldman Barrett on the brain myth that won't die, Gazzaniga on automatic brains and free people, Annaka Harris on consciousness, Zena Hitz against the cult of achievement. The questions AI made urgent were already being asked here, by people who study the original.

11 conversations here

Theories of Everything

@TheoriesofEverything

hosted by Curt Jaimungal

Jaimungal runs the only channel here where a conversation about Whitehead's process metaphysics sits next to one about AGI, and treats the adjacency as obvious rather than eccentric. Neither end gets softened to make room for the other. McGilchrist, Vervaeke, Levin, Bach, Segall: several of this collection's strangest and most load-bearing ideas arrived through here.

5 conversations here

Machine Learning Street Talk

@MachineLearningStreetTalk

hosted by Dr. Tim Scarfe

MLST is where a claim gets asked for its receipts. Scarfe holds an ML PhD and Duggar an MIT one, and they use both to push back mid-answer; the channel's own description says it covers "even highly skeptical AI positions," which is rarer than it should be. Friston on free energy, Schmidhuber on who invented what, Clune on why faster horses aren't the point — technical enough to be checkable, philosophical enough to matter.

4 conversations here

Welch Labs

@WelchLabs

Every other source here puts two people in a room. Welch Labs draws. One narrator, an animation, and a claim worked through until you can see why it holds — the compute-optimal frontier no model has crossed, the sparse autoencoders that pull nameable concepts back out of a network, LeCun's case that language was the wrong place to start. One title does this whole collection's work in six words: the moment we stopped understanding AI. That moment has a date, a paper and eight pages, and this is where you can watch it happen.

5 conversations here

StarTalk

@StarTalk

hosted by Neil deGrasse Tyson

Tyson asks the question a non-specialist would actually ask, then declines to let the specialist retreat into their own vocabulary. Astrophysics is most of the catalogue; what this collection takes is the turn inward — Donald Hoffman on whether evolution ever selected perception for accuracy, David Eagleman on why time slows in an emergency, Geoffrey Hinton on whether AI is hiding what it can do. Chuck Nice and Gary O'Reilly are there to keep it honest: an explanation that cannot survive a comedian's follow-up was not an explanation.

3 conversations here

World Science Festival

@WorldScienceFestival

hosted by Brian Greene

A physicist interviewing people whose fields are not his own — which sounds ordinary and almost never is. Greene co-founded the festival in 2008 and has spent it putting a stage under conversations that would not otherwise happen: Edvard Moser on the grid cells your brain uses to know where you are, Marcelo Gleiser on whether we invent the laws of physics or find them, Sean Carroll on what follows if many-worlds is the honest reading of quantum mechanics. The pacing is patient in the way a hard idea needs.

3 conversations here

Hannah Fry

@fryrsquared

Fry is the one who travels. Nothing in this collection comes from Fry's own channel: all five appearances were recorded somewhere else — the RSA, drawing Hassabis out on creativity; New Scientist; the Royal Institution; Bloomberg; and a TEDx stage where the talk is Fry's own. A Cambridge maths professor who asks the question the room was politely avoiding, and then doesn't accept the first answer.

Nothing drawn from their own channel — but 5 appearances in this collection, every one recorded elsewhere:

Royal Society of Arts · New Scientist · TEDx Talks · The Royal Institution · Bloomberg