It makes sense for LLMs to work with testable code for symbolic mathematics; CAS Computer Algebra System code instead of LaTeX which only roughly corresponds.
Are LLMs training on the AST parses of the symbolic expressions, or token coocurrence? What about training on the relations between code and tests?
Are LLMs training on the AST parses of the symbolic expressions, or token coocurrence? What about training on the relations between code and tests?
Benchmarks for math and physics LLMs: FrontierMath, TheoremQA, Multi SWE-bench: https://www.hackerneue.com/item?id=42097683