I have always wondered this. As more and more content on the internet becomes AI generated, does it become some kind of feedback loop where the hallucinations amplify themselves? Maybe someone smart can come along and help me understand the plan for this, if there is one.
They'll turn sites like wikipedia into free labor for training LLMs.
You are now often debugging problems directly in an LLM context window, providing back console output, error logs, etc. there is an absolute mountain of data to mine from developer chat logs.