Local AI

I think the only true way to actually stay true to privacy is using local LLM models. Even when companies like Claude or Cursor claim to not record your requests, I find it very disturbing that people will give access to their entire private code bases for these models to churn through disguised as requests. Over time even the private code bases will become training data for even better coding models. We've hit a limit on training data, so it's a big incentive for AI labs to give free access to their models for the sole reason of collecting requests as training data.