Video
Unknown · 0:00
Chain-of-thought monitoring is currently the best practical safety tool for frontier models, but it only works if reasoning stays in readable tokens; OpenAI’s leaked Astra-style architecture inserts opaque neural recu...
Unknown · 0:00
Chain-of-thought monitoring is currently the best practical safety tool for frontier models, but it only works if reasoning stays in readable tokens; OpenAI’s leaked Astra-style architecture inserts opaque neural recu...
Redirecting...