Now that SO is almost dead, how will the AI labs train their LLMs on all the programming edge cases it used to document?

Will synthetic data and documentation RAG really be enough? Or will we be stuck at 2022 debugging knowledge forever?