Whether a model's written reasoning trace accurately reflects the computation it actually used to reach its answer — not guaranteed.
Continue to AI University →