LLM self-decompilation to turn weights into conventional code

Researchers have explored a method called self-decompilation to convert large language model weights back into conventional code, aiming to better understand and analyze model internals.