Anonymize / de-identify LLM chat history export, post-processing
2 comments
This is a pretty annoying problem right now. The closest that I’ve seen is https://microsoft.github.io/presidio/
Unfortunately, it requires a decent amount of customization to do anything AFAICT
Unfortunately, it requires a decent amount of customization to do anything AFAICT
I want to conduct a study that strives to more easily anonymize the participant chat history so that when I receive it, it reduces PII risk.
another step I will need to add is just dropping chats that discuss personal health or rather summarizes chats that discuss topics of personal health? I really don't know, hence me asking here before just developing it on my own!