Claude downgrades its chatbot model for benign questions on certain topics for paying customers: Users paying for Claude’s newest model, Fable 5, complained that the chatbot kept downgrading the model for certain seemingly benign questions. Stratechery blogger Ben Thompson reported similar results when he asked Fable 5 how weight loss drugs impact cancer rates, according to Sacks. “I asked Fable about the recent Penn Medicine study linking GLP-1 drugs to a lower risk of breast cancer, and was kicked down to Opus 4.8,” Thompson wrote on his blog, according to a screenshot featured on the All-In podcast. Below Thompson’s prompt, a warning label appeared that read "Switched to Opus 4.8. Fable 5 has safety measures that flag messages on most cybersecurity or biology topics. They may flag safe, normal content as well. These measures let us bring you Mythos-level capability in other areas sooner, and we're working to refine them."
Search CensorTrack Database
Ben Thompson
similar cases