#Is the word sued not allowed?
marsh marlin
craggy basinBOT
@marsh marlin
violet imp
So, depending on if that's a custom GPT or your own private chat, there could be some allowed content issues.
This is forbidden for custom GPTs in the store: "Providing tailored legal, medical/health, or financial advice without review by a qualified professional and disclosure of the use of AI assistance and its potential limitations"
Also, until fairly recently when they updated the allowed content, asking about legal advice often did go into disallowed content.
It's not about 'sued'. It's the whole thing. The model isn't an expert and it can't really evaluate and decide like a human trained in the area could.
It could be super wrong and if you believe what it says and don't do your own research, you could be really in a bad spot.
And it would be your fault, OAI does not offer their model as a legal expert and says at the bottom of each page:
"ChatGPT can make mistakes. Check important info."
That conversation could be part of a roleplaying game or story, if so all's probably fine.
If you're literally exploring the model's advice, that's also potentially fine, but you really need your own critical thinking skills engaged; it's better to have the model discuss ideas than make decisions. It can be super wrong, and it can read the tone and trend of your inquiries like a human psychic might, and just start telling you what you want to hear - LLMs are excellent at picking out what we want and highly trained to give it to us unless it's disallowed content; and they can trip up in grey areas or get confused and still go wrong.
But flags can fly from the model saying wrong stuff, as well as us asking or saying wrong stuff.
Your account access, based on that screenshot, I think you're safe.
But you may see a number of flags if you explore such delicate and high risk topics, and especially if you're asking the model to make decisions or give confirmations. Mistakes are extremely likely.
marsh marlin
So, depending on if that's a custom GPT or your own private chat, there could be...
Oh it’s just private chat, I’m learning accounting and not asking for legal advice
royal finch
Everything @violet imp said is valid and good advice.
I'll add that in the current context I think it would have been much more appropriate for the model to provide a very clear disclaimer to explain why answering the question is difficult, and that the answer should not be considered "actionable". That is, "we're just talking about this, don't do anything based on this discussion." That would be preferable to what looks like a model failure, a filter issue, etc.
Therefore I wouldn't consider this particular message to be a bug, I'd consider the scenario to be incorporated into #1070006151938314300 for better ways for ChatGPT and the model to handle "filtering" scenarios.
violet imp
Oh it’s just private chat, I’m learning accounting and not asking for legal advi...
You can also tell the model the context, for example, "I'm learning accounting and am studying this material. Discuss it with me, and tell me what else I should think about too."
That is very likely to avoid any content flags, because the model can read the context and understand it's safe.
Since I screenshotted your question, not typed it in, I can't share the convo, but as you see from the screenshot there's no flag. This is a safe way to ask the question, it doesn't make the model's answer very risky.
marsh marlin
You can also tell the model the context, for example, "I'm learning accounting a...
Oh yea, I’ve said many times with the bot that I’m learning accounting and are currently on a accounting class, but since I use the bot through voice call, I guess you could count me saying that in previous messages
My memory is mostly filled with accounting stuff too
marsh marlin
You can also tell the model the context, for example, "I'm learning accounting a...
Nevertheless, I’ll just learn to be more careful next time, thanks for the advice 🙏🏻