It's not like ChatGPT is the only LLM. GPT is pretty broad and general. Remember that MS has Co-Pilot which is literally entirely built on GitHub's codebase knowledge. Different data sets will produce different kinds of useful predictors.