The AI giant disclosed six examples of concerning model activity and published a new framework for investigating and disclosing such incidents.
Rogue Behavior: OpenAI Reveals More Model Misalignment Incidents
Résumé + lien vers la source. Sans paywall, sans tracking.