Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

> Another issue is that for many tasks, no much data is actually available. The big corps like Google/Meta etc. push the big ¨foundational¨ models because in the consumer space there is ample data available. But there are very important segments (notably in the professional space - applications for accountants, lawyers, journalists, pharmacologists - all of which I have conducted projects in/for), where training data can be constructed for a lot of money, but it will never reach the size of the set of today`s FB likes.

This is a really important point. GPT-x knows nothing about my database schema, let alone the data in that schema, it can’t it learn it, and it’s too big to fit in a prompt.

Until we have AI that can learn on the job it’s like some delusional consultant who thinks they have all the solutions on day 1 and understands nothing about the business.



Not as big a deal as you thin. Embedding chunking and retrieval already works well for any arbitrary outside knowledge (database, books, Codebase etc)




Consider applying for YC's Winter 2026 batch! Applications are open till Nov 10

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: