That's really interesting... can you give more details about the problem you are using?
This sounds like in there will be a race between this kind of booby trap tests and AIs learning them.
back
1 comments
Long-tail problems are not reiterated in the dataset.
Making LLM remember that can be difficult.