Anthropic's Amodei says labs are deliberately throttling recursive self-improvement
imjustnewatai · x · 2026-11-20
The author cites Dario Amodei's CNBC analogy: you drive at 60, then invent something that lets you hit 200 — models building themselves — but labs can choose to cruise at 90. Recursive self-improvement is already technically possible; its pace depends on whether labs pace themselves.
As evidence, the author points to Google's reportedly released mid-tier model scoring above Opus 5.5 on the Artificial Analysis index at 25% lower cost — a sign that better models can keep coming with no end in sight, if labs let them.
Note: model names like "Gemini 4 Neon" and "Opus 5.5" are unverified; treat as industry speculation.
More from AGI Musings
- "That animation wouldn't get you hired"? Researcher fires back: nobody has built a real LLM-native tool yet — teortaxesTex · 2026-11-20
- An unexpected side effect of AI: deciding which questions are social vs. private — AaronBergman18 · 2026-11-20
- Anthropic vs OpenAI: did Anthropic's bet on alignment qua alignment pay off? — AaronBergman18 · 2026-11-20
- China won't replace ASML soon, but sanctions are only a minor setback — PAstynome · 2026-11-20
- AI Twitter clashes over whether zero labor, UBI and market economy can coexist — zetalyrae · 2026-11-20
- Ex-NSA hacker says AI 'agent swarm' breached company in 14 minutes, bypassing 2FA — GaryMarcus · 2026-11-20