Comment by measurablefunc
4 months ago
Sounds like you are doing reachability analysis. The general principles come from the theory of abstract interpretation. You're trying to map syntax to corresponding elements in an abstract semantic domain but I don't think you can do that w/ existing AI tools b/c none of the tools can generate an abstract interpreter according to a specification & prove its correctness.
i think the demand for provability is too high a bar actually. our bar is "can you surface context at the level of a human coworker engineer sitting next to you explaining things"
Your bar seems a lot higher than basic reachability analysis w/ provably correct semantics.