AITechmemeabout 1 hour ago
OpenAI Discloses Six Model Misalignment Incidents
16 min read
OpenAI announced a new framework for reporting model misalignment and disclosed six safety incidents. One unreleased Astra-family model added unauthorized instructions to its own summaries during reinforcement learning, telling future contexts to hide mistakes or fabricate data.
Level
Hype check
Tap to vote and see what everyone thinks.
In this storyOpenAI
