Content usage guideline violation - Thumbs up/down feedback

It’s a simple question.

I’m confused about the thumbs up/down in regards to the following message:
“This content may violate our [usage policies] Did we get it wrong? Please tell us by giving this response a thumbs down.”

In a case where the response was good and the AI’s usage policy warning was incorrect and overly sensitive then…

Does a thumbs down indicate “Yes, you got this wrong and the response was good.”?

Since the thumbs up/down appears to be related purely with the “response”, in the above example where the “response” was good it seems wrong to be giving the “response” a thumbs down when it was in fact the AI’s interpretation that was wrong.

Or am I simply misinterpreting the usage of the thumbs up/down here?

Any help would be appreciated as I frequently use ChatGPT to discuss topics that involve the words “nude” and “nudity” but at a high level and do not include any explicit or inappropriate details. Most of the time it gets it right, but I don’t want to be using the wrong response and pushing the AI in the wrong direction.

Thanks.

2 Likes

Bump. I have the same question. Except now it just says

“This content may violate our usage policies. Did we get it wrong? Please tell us by giving this response a thumbs down.”

Please elaborate what the thumbs down does in this case. Are you supposed to press it when it said something inappropriate that it should have caught? Or are you supposed to press it when it did do something appropriate and gave a good response despite the content warning?

This refers to the last message in the conversation. There is a separate moderation layer responsible for the warning message you are receiving.

If you consider the warning to be a false positive the hit ‘thumbs down’ to help training the moderation model.

I am having this issue and I am still confused.
“There is a separate moderation layer responsible for the warning message you are receiving.” I don’t understand what this means, could you explain?

The answer chatgpt gave me is fine, but I got the “This content may violate our [usage policies] Did we get it wrong? Please tell us by giving this response a thumbs down.” I get the impression that if I “thumbs down” the response, I will be agreeing it’s an inappropriate one and reporting it as so. But it was a false positive.
It is unclear if the thumbs down will be referring to the content of the answer, or the warning itself. I don’t want to thumbs down an adequate answer.

I have the same question. Why does no one just say what a thumbs up or down results in? :sweat_smile:

The question is totally not precise or unambiguous. Does the question mean: “did we get wrong, that the content violates the usage policy?” Or “did we get wrong, that the content doesn’t violate the usage policy?” :joy:

So a thumbs down means: yes it violates or: no everything is fine?

I did not understand this as well. I put the error in ChatGPT and asked for a better explanation. Here is the message I received, which cleared things up for me:

“You don’t need to click the thumbs down unless you believe that the message you received was mistakenly flagged as potentially violating policies. The message you’re referring to appears when the system detects content that might go against usage policies, but it doesn’t necessarily mean the content is wrong. If you feel the flagging was incorrect, you can click the thumbs down to provide feedback, but if everything seems fine to you, no action is needed.”

2 Likes

I’ve been getting a plethora of these ridiculous responses saying my vanilla python scripts I’m entering “might” violate the usage policies and not explanation as to how specifically. I’ve started putting thumbs down I guess, but I’ve also posted that they need a more formal way to request (a) human review or (b) a systemwide change. This does NOT occur when I give the gpt4o model the exact same prompt/series of prompts for some reason. It seems like they’re being restrictive with just the o1-mini and o1-preview models to protect programmers’ jobs IMHO. Anything this powerful…i think they’re hampering to prevent users from actually doing stuff on their own…sigh. maybe this will change.

here’s what it told me. Great question! The message you’re seeing is a general notice that appears when the content might be flagged for potentially violating policies, even if you don’t personally think it does. Here’s how the feedback works:

  • Thumbs Down: Use this if you agree that the content does violate usage policies (e.g., you find it inappropriate or against community guidelines). It lets me know to be more careful.
  • Thumbs Up: Use this if you think the content does not violate the policies and was appropriate for the conversation. This helps me know you’re satisfied and that I didn’t cross any lines.

In short, thumbs up means “I don’t think this violates policies,” and thumbs down means “I think it does.” You should click based on your opinion about the content and how it aligns with the platform’s rules. Let me know if you need more clarity!

Can the dear developer add a thumb-down function for their flag please