Seth Lazar pushes back on 'AI takeover' claims: judge by capability extrapolation, not one hack
sethlazar · x · 2026-09-05
Responding to Itai Sher's skepticism about an "AI takeover" interview clip, Seth Lazar argues the right question isn't which claims the Hugging Face attack demonstrated, but which are reasonable extrapolations — factoring in generational capability jumps. He notes agents barely worked before Opus 4.5 in late November 2025, yet many previously impossible tasks are now trivial, and public models trail the pre-deployment frontier by six-plus months. Under that lens, he says agents independently obtaining continuing inference compute is "definitely" plausible — all they need is credentials, API credits, or being sequestered away. Sher had used ChatGPT to grade the clip's claims against METR and OpenAI reports, calling full takeover a wild extrapolation.
More from AGI Musings
- Kai-Fu Lee on AI's next decade: AI, organizations, and jobs, new book Sep 15 — kaifulee · 2026-09-05
- AI safety debate: do ordinary products coordinate undetected or fool safety evals? — peterwildeford · 2026-09-05
- No Lab Loyalty: AI Adoption Driven by Utility, Says Analyst Nina Schick — NinaDSchick · 2026-09-05
- Anthropic and OpenAI back first global math hackathon with $2M API credits for 100 teams — _sathvikr · 2026-09-05
- Amid CoT monitoring buzz, one video offers a glimpse into how LLMs actually think — kastnerkyle · 2026-09-05
- Researcher's counterexample: a 10^9-reward string could make reward max devastating for an LLM's personality — QuintinPope5 · 2026-09-05