“A primary reason OpenAI stopped showing chain-of-thought reasoning is to preserve it as a faithful interpretability mechanism, avoiding the temptation to train the model to produce reasoning that looks good to users but isn't authentic.”