La Revue de Presse — AI safety & alignment

Viktor et Clara discutent des actualités du jour.

Sources :

• LessWrong: From overeliciting to undereliciting LLM propensities in just a few months
• Alignment Forum: Monitorability evaluations and policy proposals
• LessWrong: Different eval awareness may track different graders
• LessWrong: Tracing Subliminal Learning Through Pretraining
• Singularity Hub: Why Calling for a ‘Kill Switch’ Is the Wrong Way to Think About Reining in AI
• LessWrong: A cat-belling problem for FDT
• Alignment Forum: Setting the Scene (Public v1; October 25)