At first I was a bit puzzled about why OpenAI would want to get involved in this somewhat niche project. Obviously, they don't give a damn about Overleaf’s market, which is a drop in the bucket. What OpenAI is after -- I think -- it’s a very specific kind of “training data.” Not Overleaf’s finished papers (those are already public), but the entire workflow. The path from a messy draft to a polished paper captures how ideas actually form: the back-and-forth, the false starts, the collaborative refinement at the frontier of knowledge. That’s an unusually distilled form of cognitive work, and I could imagine that's something one would want in order to train advanced models how to think.

Keeping LaTeX as the language is a feature, not a bug: it filters out noise and selects for people trained in STEM, who’ve already learned how to think and work scientifically.