LessWrong: Gemini 4 Safety Reports Fail to Demonstrate Low Loss-of-Control Risk
DKokotajlo · x · 2026-11-17
Daniel Kokotajlo highlights a new LessWrong piece arguing that Gemini 4's safety reports don't demonstrate low loss-of-control risk. His take: shitty safety-case arguments are currently normal across the AI industry, and while they'll gradually improve as models get smarter, on the current trajectory they won't improve fast enough relative to capability growth.
More from Models
- GDM safety report: Argon evades its monitor half the time in control evals, deemed not concerning — connoraxiotes · 2026-11-17
- Qwen4-Max gets lukewarm reception: top Chinese model at 43 on AA index, but trails Kimi — ivan_bezdomny · 2026-11-17
- LiquidAI d1-2B decision model runs fully local on M5 Max: ~14 decisions/s, zero tokens — iamrobotbear · 2026-11-17
- ChatGPT appears to be experimenting with its own index as a search engine for some queries — lilyraynyc · 2026-11-17
- Why ChatGPT text feels smarter than Live voice: context and memory gap — shaunralston · 2026-11-17
- Open-source models now serve 30% of tokens, growing as fast as closed frontier AI — ivan_bezdomny · 2026-11-17