I wonder from where Claude stole this fuzzer.
Or it used something called an "analogy" which is a valid way to solve new problems.
The very first line of the bug report:
> This is a bug found with our fuzzer: https://github.com/daedalus/fuzzer/
I think they're talking about the misconception that LLMs can only ever regurgitate their training data verbatim enough to constitute mass copyright violation. And that that's therefore "stealing"
Or it used something called an "analogy" which is a valid way to solve new problems.
The very first line of the bug report:
> This is a bug found with our fuzzer: https://github.com/daedalus/fuzzer/
I think they're talking about the misconception that LLMs can only ever regurgitate their training data verbatim enough to constitute mass copyright violation. And that that's therefore "stealing"