I can give examples as seems useful.
Also, AIs—at least Claude—sometimes silently sandbag on other topics (intentionally producing a poor or low detail answer without making this clear). This is likely an unintended generalization. (Anthropic's consitition clearly prohibits this.) Companies should fix this.
See also the thread here:
@RyanPGreenblatt Silent sandbagging is fine! All models already silently sandbag adjacent to blocked topics, like microbiology, writing erotica, fantasy villains'evil plans. Intentional silent sandbagging on a fraction of traffic comparable to those topics topics doesn't change anything.
Also, I think silent sandbagging is a decent amount less bad when not done at the AI layer and when instead done at a somewhat higher layer (like Anthropic originally did for AI R&D with fable). It still seems bad.
See also the thread here:
I can give examples as seems useful.
Also, AIs—at least Claude—sometimes silently sandbag on other topics (intentionally producing a poor or low detail answer without making this clear). This is likely an unintended generalization. (Anthropic's consitition clearly prohibits this.) Companies should fix this.