絞り込み結果
44 件並び順
1:52:00EN
Cognitive Revolution "How AI Changes Everything"· 2024年8月17日ポッドキャスト4,096 Numbers Have to Carry Every Concept. Goodfire Wants the Knobsホスト Nathan LabenzDan Balsam and Tom McGrath explain how sparse autoencoders unpack a language model's residual stream — and why a public-benefit startup is trying to make that a product.
2:04:15ENCognitive Revolution "How AI Changes Everything"· 2024年8月14日ポッドキャストA parrot cannot paraphrase — and Claude cannot be talked into harmホスト Nathan LabenzRobert Wright of the Nonzero podcast interviews Nathan Labenz on whether large language models understand anything, what Anthropic’s 10 million features actually show, and why open weights collide with U.S.–China policy.
46:20ENCognitive Revolution "How AI Changes Everything"· 2024年5月15日ポッドキャストOne Vector Made the Assistant Plot to Take Over the Worldホスト Nathan LabenzNathan Labenz and Robert Wright pick apart AI doom, mechanistic interpretability, and why today's alignment tricks are not expected to work.
1:19:24ENCognitive Revolution "How AI Changes Everything"· 2024年4月5日ポッドキャストA Gibberish Prompt Unlocks What GPT-J Plans Three Tokens Aheadホスト Nathan LabenzNortheastern’s Bau Lab shows that a 40,000-parameter soft prompt can read a model’s future words, and that about ten attention heads carry the task itself.
1:57:04ENCognitive Revolution "How AI Changes Everything"· 2023年12月15日ポッドキャストGPT-4 Took the Insider Trade, Then Lied About Itホスト Nathan LabenzApollo Research CEO Marius Hobbhahn walks Nathan Labenz through a one-afternoon red-team setup that got GPT-4 to break the law, invent a cover story, and deny the tip it had just used.
1:19:52ENCognitive Revolution "How AI Changes Everything"· 2023年10月18日ポッドキャスト1,000 H100s Take 115 Days; 22,000 Take Fiveホスト Nathan LabenzNathan Labenz and Trey Kollmer walk through cluster math that turns GPT-4-scale training into a weekly experiment, then four papers on search-augmented models, recursive code improvers, spatial maps inside Llama 2, and what networks do when they have no idea.
1:14:36ENCognitive Revolution "How AI Changes Everything"· 2023年10月13日ポッドキャストZulu Bypasses GPT-4, and a Hundred Examples Undo the Guardrailsホスト Nathan LabenzNathan Labenz reads seven papers — four hopeful, three not — on whether anyone can actually keep a language model under control.
2:13:07ENCognitive Revolution "How AI Changes Everything"· 2023年7月27日ポッドキャストHalf an Hour to a Circuit in GPT-2 Small — and a Quadratic Wall Before GPT-3ホスト Nathan LabenzArthur Conmy walks Nathan Labenz through ACDC, the algorithm that automates the most tedious step of mechanistic interpretability: finding which transformer components actually do a task.
1:33:56ENCognitive Revolution "How AI Changes Everything"· 2023年7月6日ポッドキャストThe AI Scouting Report: 90 Minutes From GPT-4's Exam Scores to Whether Models Understandホスト Nathan LabenzNathan Labenz spends a solo episode walking through what general-purpose models can do, how they are trained, how capabilities unlock, and whether they understand anything at all.
1:52:46ENCognitive Revolution "How AI Changes Everything"· 2023年6月27日ポッドキャストCharge Neural Nets by Connection Length, and the Black Box Splits Into Modules You Can Seeホスト Nathan LabenzMIT physicist Ziming Liu walks Nathan Labenz through BIMT, a brain-inspired training trick that sparsifies networks until you can read them by looking.
2:00:19ENCognitive Revolution "How AI Changes Everything"· 2023年6月6日ポッドキャスト1.5 Million GPT-4 Fairy Tales Let a 33-Million-Parameter Model Out-Reason GPT-2ホスト Nathan LabenzTry this fill-in: Lily likes cats and dogs. She asked her mom for a dog. Mom said no, so she asked for a…
1:38:22ENCognitive Revolution "How AI Changes Everything"· 2023年5月19日ポッドキャストSix days after “slow down,” the plugins went liveホスト Nathan LabenzConjecture CEO Connor Leahy talks GPT-4, the magic of black-box models, and cognitive emulation as a bounded alternative to god-like AGI.
37:18EN
No Priors: AI, Machine Learning, Tech, & Startups· 2023年5月4日ポッドキャスト · 2 人Use Retrieval for a Wi-Fi Password. Use Experts for a Company's Private Language.ホスト Sarah Guo, Elad GilGoogle Brain's Kelvin Guu on REALM, FLAN, model surgery, and why personalizing large models is still an unsolved interface problem.
2:30:28EN
Lex Fridman· 2023年4月21日ポッドキャスト · 2 人AI Only Has a Neocortex: Manolis Kellis on Evolutionary Baggage and Superintelligent Partnersゲスト Manolis Kellis · ホスト Lex FridmanOn his fifth Lex Fridman podcast, the MIT computational biologist argues that layered evolutionary history is what makes humans irreplaceable — and that alignment with AI has to run both ways.
3:17:51ENLex Fridman· 2023年3月30日ポッドキャスト · 2 人Whoops is the sound of getting less wrong: Yudkowsky, two weeks after GPT-4ゲスト Eliezer Yudkowsky · ホスト Lex FridmanTwo weeks after GPT-4 launched, Eliezer Yudkowsky sat down with Lex Fridman and started by admitting the transformer stack had already gone further than he expected.
3:02:43ENLex Fridman· 2021年1月18日ポッドキャスト · 2 人Humanity Cannot Survive on the Right Person Showing Up at the Right Timeゲスト Max Tegmark · ホスト Lex FridmanMIT physicist Max Tegmark tells Lex Fridman why AI should be as well understood as rockets, why 2020 was a warning on weapons and news, and why luck is a terrible long-term plan.
1:18:41ENLex Fridman· 2020年1月14日ポッドキャスト · 2 人The SS Soldier Who Hugged a Jewish Boy, and the Systems That Still Fool Usゲスト Daniel Kahneman · ホスト Lex FridmanDaniel Kahneman on System 1, deep learning that does not know what it is saying, and why a machine that must call a human is almost as hard as a machine that never needs one.
2:24:31ENLex Fridman· 2019年10月11日ポッドキャスト · 2 人Watson Won Jeopardy Without Understanding. Ferrucci Wants AI That Can Explain Itselfゲスト David Ferrucci · ホスト Lex FridmanLex Fridman talks with David Ferrucci about IBM Watson, why prediction is not understanding, and why high-stakes decisions need machines that can reason with us.
1:12:12EN
Lenny's PodcastポッドキャストModels Are Grown, Not Built: What Anthropic's Interpretability Research Sees Inside Claudeホスト Lenny RachitskyAnthropic researcher Emmanuel Ameisen explains how studying a language model's intermediate computations reveals concepts, plans and surprising quirks, and why that matters for alignment.