Tell HN: Fable guardrails trigger on random questions
I have Fable guardrails trigger on seemingly innocuous questions. Two examples: a) What is Echium b)I want you to deep dive on coffee and it’s effect on health, cognition and longevity. The output should be something that tells you areas where the evidence is strong and where it is weak but plausible and where there is no strong evidence. Does anyone know why this happens? I have many such examples where it will randomly switch models.
1 comments
I think Anthropic made the guardrails overly restrictive on purpose, considering it was unprecedented how powerful the model was. I think it was also intended to help their PR, as they've always marketed themselves as the "ethical" AI lab, so they wish to go overboard so people think they're just working hard on their filters (which I don't doubt they are), especially when we look at the money behind it.