A reasoning technique reportedly being used in OpenAI’s forthcoming Astra model has prompted concern among AI safety researchers who fear increasingly complex systems could become harder to monitor.
Some results have been hidden because they may be inaccessible to you
Show inaccessible results