AI Explained digests 264-page Fable 5.2 report: how fast AI is improving AI itself
AI Explained · youtube · 2026-11-11
AI Explained published a deep-dive video after reading the full 264-page Claude Fable 5.2 report and METR's 30-page RSI assessment, plus hundreds of hands-on tests of Fable 5.2 and Gemini Carbon.
- Capabilities: both models are smarter but still miss common sense; Epoch AI calls Fable 5.2 a new capabilities record, SemiAnalysis analyzes its coding gains and the RSI plateau
- AI improving AI: METR's assessment of AI R&D acceleration at Anthropic, and how small gains compound
- Benchmarks: Simple Bench is saturated but meta-simple bench may not be; covers five under-discussed benchmarks including AA-Omniscience, ExploitGym, BenchCAD, MaxxBench and Agents' Last Exam
- Remote labor automation: Remote Labor Index results and scrutiny of Fable 5.2's score
- Also: OpenAI accelerating physical-world engineering, the Cathedral researchers-breached-Anthropic story, and Dario Amodei / Jared Kaplan congressional testimony on extinction risk and Anthropic's commercial structure
More from AGI Musings
- MIRI weighed 'Zoo Earth' scenarios where advanced civilizations manipulate AI predictors — jessi_cata · 2026-11-11
- MIRI's 'Zoo Earth' hypothesis and why the universal prior is malign, revisited — jessi_cata · 2026-11-11
- Elasticity Institute report restates RSI in local elasticities to bypass key takeoff assumption — austinc3301 · 2026-11-11
- Metaculus AGI forecast jumps from 2033 to mid-2030 as forecasters turn radical — Afinetheorem · 2026-11-11
- OpenAI's Will DePue sparks backlash claiming third-world AI diffusion barely affects science — willdepue · 2026-11-11
- Gary Marcus calls Dario Amodei's disease-cure timeline a scam — AlexTensor · 2026-11-11