‘We may release models we can’t control’: Amodei, Altman … and Gemini Carbon
From AI Explained · · 27 min

The player loads from YouTube only when you click. It uses YouTube's privacy-enhanced mode.
If the video does not play, the owner may have turned off embedding. Watch it on YouTube instead.
Description
You know what, by AI standards, it might not have been the craziest week, but that doesn’t mean several headline-worthy things happened - from Altman and Anthropic implying they would release models they can’t control, with behavior they cannot verify, to Gemini 4 Carbon bursting onto the scene, some Amodei rationalization, de-tokenization and more…
AI Insiders ($9!): https://www.patreon.com/AIExplained
Chapters: 00:00 - Introduction 00:00 - Introduction 01:26 - Anthropic Staff Survey 04:32 - Amodei's Five Rationalisations 11:43 - Altman: Only Alignment Will Work 18:41 - Gemini 4 Carbon and Why Models Keep Impressing
Anthropic: The AI expectations gap: https://www.anthropic.com/research/ai-expectations-gap
Dario Amodei responds to Nathan Calvin: https://x.com/DarioAmodei/status/2112732230124818891
Anthropic Responsible Scaling Policy v3.0: https://www-cdn.anthropic.com/e670587677525f28df69b59e5fb4c22cc5461a17.pdf
Responsible Scaling Policy v3 discussion: https://www.lesswrong.com/posts/HzKuzrKfaDJvQqmjh/responsible-scaling-policy-v3
Anthropic's core views on AI safety: https://www.anthropic.com/news/core-views-on-ai-safety
Nathan Calvin on Anthropic's IPO risk disclosures: https://x.com/_NathanCalvin/status/2114720411413819410/photo/1
OpenAI: Securing frontier model training: https://openai.com/index/securing-our-frontier-model-training-environment/
Google: Introducing Gemini 4 Carbon: https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/
SemiAnalysis: GDPx benchmark: https://x.com/SemiAnalysis_/status/2117075900575477855
The Information: OpenAI's next flagship and safety issues: https://www.theinformation.com/articles/openai-scrambles-fix-safety-issues-shipping-next-flagship-model
AI Explained: Understanding Models — Impossible, Solved, or Not Needed?: https://www.patreon.com/AIExplained/posts/exclusive-models-172917747
Dario Amodei: We Must Pace the Frontier: https://darioamodei.com/post/we-must-pace-the-frontier
Dario Amodei: The Adolescence of Technology: https://darioamodei.com/essay/the-adolescence-of-technology
Evan Hubinger on recursive self-improvement risks: https://x.com/EvanHub/status/2097528891846074828
Leo Gao on AI lab employees' shrinking influence: https://x.com/nabla_theta/status/2115922100204810594
Jack Clark on superintelligence risks, BBC interview: https://x.com/aleshu_ber/status/2116988191597060409
Jack Clark: Import AI: https://jack-clark.net/
Dylan Patel's apology to Google over Gemini Carbon: https://x.com/dylan522p/status/2117092550842683671
SemiAnalysis: Gemini Is Cooked but GCP Is Cooking: https://newsletter.semianalysis.com/p/gemini-is-cooked-but-gcp-is-cooking
Vivek Natarajan on Gemini Carbon and RSI: https://x.com/vivnat/status/2115497643216707621
Albert Gu on learning directly from raw data: https://x.com/_albertgu/status/2115502110578290900
Garrison Lovely: Screenshots of Tony Feng's model claims: https://x.com/GarrisonLovely/status/2116226357101199369/photo/1
Jason Clinton on worsening cyber threats: https://x.com/JasonDClinton/status/2113325263157395456
Jason Clinton: "We tried to warn the world": https://x.com/JasonDClinton/status/2113363809629020374
Mustafa Suleyman: Don't build conscious superintelligence: https://x.com/mustafasuleyman/status/2116913722878820528
Stockfish chess engine: https://en.wikipedia.org/wiki/Stockfish_(chess)
Non-hype Newsletter: https://signaltonoise.beehiiv.com/