Google DeepMind withholds reward optimization from internal chain-of-thought tokens to keep the reasoning process unfiltered for safety monitoring

Activity

Activity (0)
Loading comments...
Opening sign-in…