AI Explained: Fable 5.2: The Self-Improvement Loop Keeps Loopin’ … (plus Gemini Carbon)

I’ve read the full Fable 5.2 report at 264 pages, the 30 page METR RSI report, and tested Gemini Carbon and Fable 5.2 hundreds of times. There’s too much to cover in one video, but let’s explore everything from how fast AI is accelerating AI, five benchmarks few others will talk about, why Simple Bench got saturated but meta-simple bench might not have, Internal model 4, remote labour automation and much more



https://www.assemblyai.com/aiexplained



Exclusive Vids($9!): https://www.patreon.com/AIExplained



00:00 - Introduction

00:49 - how good are Fable 5.2 and Carbon?

09:01 - smarter, but still missing some common sense

15:49 - how fast is AI improving itself?

23:51 - small gains, big consequences

26:33 - it knows when you're watching







Introducing Claude Fable 5.2 and Mythos 5.2: https://www.anthropic.com/claude-fable-and-mythos-5-2



Claude Fable 5.2 and Mythos 5.2 System Card: https://assets.claude.com/27d60918d218125e7504bc7354c4b4626ec56b11.pdf



METR: AI R&D acceleration at Anthropic: https://metr.org/blog/2026-11-05-anthropic-ai-rd-acceleration-assessment/



Epoch AI: Fable 5.2 sets a new capabilities record: https://x.com/EpochAIResearch/status/2119903415363113088



SemiAnalysis: Fable 5.2's coding gains and RSI plateau: https://x.com/SemiAnalysis_/status/2119961452496756926



SemiAnalysis GDPx: https://gdpx.semianalysis.com/



Remote Labor Index: https://www.remotelabor.ai/



Remote Labor Index paper: https://arxiv.org/pdf/2510.26787



Greg Burnham on Fable 5.2's Remote Labor Index score: https://x.com/GregHBurnham/status/2120155251240292619/photo/1



Agents' Last Exam leaderboard: https://agents-last-exam.org/leaderboard



AA-Omniscience: Knowledge and hallucination benchmark: https://artificialanalysis.ai/evaluations/omniscience



ExploitGym leaderboard: https://benchlm.ai/benchmarks/exploitgym



ExploitGym: Turning vulnerabilities into real attacks: https://arxiv.org/pdf/2605.11086



BenchCAD leaderboard: https://benchcad.com/leaderboard



BenchCAD paper: https://arxiv.org/pdf/2605.10865



OpenAI: Accelerating engineering for the physical world: https://openai.com/index/accelerating-engineering-for-the-physical-world/



Google: Gemini 4 Carbon is now available: https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-carbon-now-available/



Justin Lo: MaxxBench tests coding robustness: https://x.com/loresjpg/status/2118880329490637049/photo/1



Claude: Fable 5.2 and Mythos 5.2 capabilities and speed: https://x.com/claudeai/status/2119867117642146270/photo/1



Washington Post: Cathedral researchers got inside Anthropic: https://www.washingtonpost.com/technology/2026/10/26/researchers-cybersecurity-start-up-cathedral-got-inside-anthropic/



Max Winga: Dario Amodei's testimony on AI extinction risk: https://x.com/maxwinga/status/2119034229983986049/photo/2



Jared Kaplan's testimony on Anthropic's commercial structure: https://x.com/_smitop/status/2119870258919969066/photo/2



Anthropic: AI-driven R&D risk disclosure excerpt: https://pbs.twimg.com/media/HWxojYda4AAu5c9?format=jpg&name=900x900



Mrinank Sharma's Anthropic resignation letter: https://x.com/MrinankSharma/status/2020881722003583421/photo/1



Chris Olah on Pope Leo XIV's encyclical: https://www.anthropic.com/news/chris-olah-pope-leo-encyclical



LessWrong: Claude 3's launch and GPT-4 performance claims: https://www.lesswrong.com/posts/JbE7KynwshwkXPJAJ/anthropic-release-claude-3-claims-gpt-4-performance



Alex Albert: Bringing Leonardo da Vinci's notebooks to life: https://x.com/alexalbert__/status/2119916954337349737



Claude: Watching subagents work inside a project: https://x.com/claudeai/status/2120174853856821549



IntegrityBench: https://integrity-bench.com/



SimpleBench: https://simple-bench.com/









Non-hype Newsletter: https://signaltonoise.beehiiv.com/



Podcast: https://aiexplainedopodcast.buzzsprout.com/