A lot of the training data is in markdown— documentation, GitHub readme.md files, etc. Also it’s not just about putting it in markdown, it’s also just quickly putting all the text in a single place along with the file paths so the model can better understand how all the pieces fit together.
reply