john teichmanEventsAnti-Scheming: how would we train a model not to scheme? (Bronson Schoen and Marius Hobbhahn, Apollo Research) 2025-11-08