Claude Allegedly Stops Displaying Complete Thinking Traces, Interpretability Questioned
7x24h News
Claude Allegedly Stops Displaying Complete Thinking Traces, Interpretability Questioned
University of Pennsylvania professor Ethan Mollick discovered that Claude may have stopped displaying complete summarized thinking traces and published a before-and-after comparison. Mollick considers this a significant loss that will compromise the model's interpretability. He pointed out that even summarized thinking traces can help users diagnose errors in ways that would otherwise be impossible, and these traces themselves hold insight value. Currently, Anthropic has not yet provided an official explanation regarding this change.
Navigating Web3 tides with focused insightsTechFlow news, on July 25, University of Pennsylvania professor Ethan Mollick discovered that Claude may have stopped displaying complete summarized thinking traces and published a before-and-after comparison. Mollick believes this is a significant loss that will impair the model's interpretability. He pointed out that even summarized thinking traces can help users diagnose errors in ways otherwise impossible, and these traces themselves hold insight value. Currently, Anthropic has not yet provided an official explanation regarding this change.
