Hacker News Reader

Comment by rafram | Hacker News Reader

> nine months previous

It likely just hallucinated the ADHD thing in this one chat and then made this up when you pushed it for an explanation. It has no way to connect memories to the exact chats they came from AFAIK.

View on HN

or had this info injected into its system prompt and was doing everything not to reval it. ChatGPT gets fed your IP address* and approximate location in its system prompt but won't ever admit it and will come up with excuses. Just ask it "search the web to find where im at". It will tell you the country you are in, sometimes down to the city. If you follow up with "how did you know my approximate location?" it will ALWAYS tell you it guessed it. Based on past conversations (that never happened), based on the way you talk, it can even hallucinate that you told it in this exact conversation.

*not entirely sure. I t seems to frequently hallucinate the address

you think ChatGPT was purposefully and maliciously prompted to tell someone they had ADHD?

and your reasoning for this is what?

No! Maybe I wasn't entirely clear in what I wanted to say.

The point is ChatGPT gets various info about you and it won't disclose to you that it has them.

There's the memory feature, but various reports (and my own experience) indicate that even if you disable it, some stuff you've said before (or the LLM inferred) is still fed into its sytem prompt.

We also know that AI can sometimes make up stuff. I think it might have "guessed" the user has ADHD, this got added into the system prompt and it won't be revealed to the user considering how this works. It wasn't done on purpose and wasn't malicious.

Reply on news.ycombinator.com