Liron Shapira on Catastrophic AI Risks | Podcast #7 скачать в хорошем качестве

Liron Shapira on Catastrophic AI Risks | Podcast #7

Liron Shapira is entrepreneur, an angel investor, and founder/CEO of Relationship Hero. Liron has been one of the most outspoken voices on AI risk and dedicates his popular YouTube Show, Doom Debates, to debating experts on the risks of super-intelligent AI systems. Chapters 00:00:00 – Intro & who Liron Shapira is (Doom Debates, background) 00:02:30 – Bayesian mindset, thought experiments & “knowing without perfect tests” (JWST example) 00:04:12 – EA & rationalist communities; LessWrong, CFAR, overlap with AI safety 00:06:06 – Four AI risk buckets; why “rogue AI” worries him most 00:07:39 – Could there be a dystopian interlude before “game over”? 00:08:49 – Timelines & Yudkowsky’s If Anyone Builds It, Everyone Dies (conditional prediction) 00:10:16 – What “P(doom)” means; Liron’s rough 50% estimate 00:16:16 – Why this is a serious debate (CAIS statement & signatories) 00:19:19 – Nuclear weapons analogy: discovering the “game-over button” 00:26:55 – US vs EU/UK concern levels; “AI as a space race” & David Sacks point 00:28:56 – Can AI be controlled? Epistemology, “meta-knowledge,” and 50/50 optimism long-run 00:37:55 – “LLMs are just cultural mirrors”? Benchmarks & coding superhumanity rebuttal 00:39:23 – Orthogonality thesis: intelligence vs values aren’t the same axis 00:41:19 – Instrumental sub-goals: why even “good” AIs could exterminate (for other ends) 00:43:00 – Penrose & quantum brains: why that critique doesn’t land today 00:45:43 – Narrow AI, “pause” intuition & flying too close to the sun (Icarus/shuffleboard) 00:48:08 – The real dangerous ingredient: outcome-to-action planning (not “agency” per se) Liron's Youtube Chanel : ‪@DoomDebates‬