The math story is fascinating, but honestly the data-ownership issue feels like the bigger takeaway. If researchers can spend a year developing an unpublished result while their AI-assisted work accumulates inside one provider’s systems, it creates a pretty uncomfortable dependency—even if nobody did anything wrong. Using multiple models helps, but controlling where the working state and source material live seems like the much more important part.
The math story is fascinating, but honestly the data-ownership issue feels like the bigger takeaway. If researchers can spend a year developing an unpublished result while their AI-assisted work accumulates inside one provider’s systems, it creates a pretty uncomfortable dependency—even if nobody did anything wrong. Using multiple models helps, but controlling where the working state and source material live seems like the much more important part.