Loading activity…
@danielchtan97
alignment research lead @arcadiaimpact. prefer email to DM (unless mutuals) ai labs should slow down / pause
AI
The research team says it found new failure modes and implementation details critical to good performance, and is open-sourcing its work.
AI
AI safety researcher Daniel Tan posted that alignment might have been solved without reinforcement learning.
AI
Researcher highlights Redwood demonstration separating reward hacking from misalignment.
