Fuzzing Ruby to finding invalid code

[enebo]

I said the other day it could take year(s) to find all the invalid programs that should not be parsed because we do not have a repository of invalid ruby code snippets. I realized if we made a fuzzer we could probably generate this data and get a lot of cases figured out quickly. How hard would this be? Could a LLM possibly be used too?