Solving Problems in AI: LLM Explainability, Hallucinations and More [video]

The first speaker in the video works at or started https://neocortix.com/ where, by "doing surgery" on LLAMA, he is able to have the AI recognize when it's reaching for facts, then find the actual reference in the training data set and quote it directly.

I'd have linked the site instead of a video, but the site doesn't have too much info while in the video he shows a grokkable breakdown of how it works.