the base model is just predicting the likely continuation of the prompt. and it’s a reasonable prediction that, when an assistant is given a harmful instruction, they will refuse. this behaviour isn’t surprising.
This is not an obvious continuation of the prompt to me—maybe there are just a lot more examples of explicit refusal on the internet than there are in (e.g.) real life.
My current best guess for why base models refuse so much is that “Sorry, I can’t help with that. I don’t know how to” is actually extremely common on the internet, based on discussion with Achyuta Rajaram on twitter: https://x.com/ArthurConmy/status/1840514842098106527
This fits with our observations about how frequently LLaMA-1 performs incompetent refusal
Is there a reason to expect this kind of behaviour to appear from base models with no fine-tuning?
the base model is just predicting the likely continuation of the prompt. and it’s a reasonable prediction that, when an assistant is given a harmful instruction, they will refuse. this behaviour isn’t surprising.
This is not an obvious continuation of the prompt to me—maybe there are just a lot more examples of explicit refusal on the internet than there are in (e.g.) real life.
My current best guess for why base models refuse so much is that “Sorry, I can’t help with that. I don’t know how to” is actually extremely common on the internet, based on discussion with Achyuta Rajaram on twitter: https://x.com/ArthurConmy/status/1840514842098106527
This fits with our observations about how frequently LLaMA-1 performs incompetent refusal