Claude is suspected to have stopped showing a complete picture of his thoughts

On July 25, a professor at the University of Pennsylvania, Ethan Mollick, found that Claude may have stopped showing complete summary thought marks and published a comparison. Mollick considered this to be a significant loss that would undermine the interpretability of the model. He pointed out that even the summarised traces of thought could help users to diagnose errors in ways that otherwise would not be possible, and that they were per se of invisible value. Anthropic has not yet explained this change。