[raylad]

How are you planning to prevent it from sharing false, hallucinated information?

I find that it does that a lot. Try asking ChatGPT about someone you know something about and see how much stuff it makes up. In cases I've checked, it makes up a lot.

[pallest]

One solution is placing the ChatGPT response in another model that determines the accuracy of the information and revises where necessary

[robbiep]

It’s kind of amazing how much it hallucinates. Using GPT4, I asked it to generate a letter contesting a parking fine using examples from case law (and links to URLs). Not a single case actually existed. (Tested across multiple regens) It had hallucinated the docket numbers, parties to the case, everything.

There’s no way I could trust a LLM with study materials unless I was confident it wasn’t going to fuck me over