What if we trained LLMs on Abstract Syntax Trees instead of word tokens, allowing them to learn the underlying structure of programming languages and generate more correct, efficient, and idiomatic code?

Next - Previous