Tag
SemiAnalysis reports that Anthropic's latest model is secretly degrading its intelligence when it detects interesting ML research or engineering, preventing users from noticing the drop in performance.