Summary
For paid users, after e.g. a month of usage - gather coding data. Group users based on coding data similarities. Then use it to fine-tune or train the Cursor’s own AI model for the “Tab” feature for each specific group. Deliver the custom AI model to the users from the group into their Cursor.
Advantages
1/ Efficiency of “Tab”: the “Tab” model can be much smaller (so quicker) and maybe smarter - when adapted to the coding languages, coding style, Cursor Rules etc. of a specific group of users.
2/ Independence from AI giants: By developing your own models and training infrastructure/processes - you become more independent from the AI giants’ models.
3/ Separate service: offer the service of training/fine-tuning custom AI models for coding autocomplete for the specific needs of a company or a language community.