L'Analyse Longue — AI safety & alignment

Clara et Nia discutent des actualités du jour.

Sources :

• Alignment Forum: The Case for Model Forensics
• LessWrong: Deployment Awareness Matters More Than Evaluation Awareness
• 3 Quarks Daily: We Need an International Treaty to Ban Superintelligence
• LessWrong: Some subtypes of taskishness / corrigibility