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.
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?
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?
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?”
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.”
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!
I agree, the language is confusing and could cause people to accidentally confirm false flags. I initially interpreted it as “if the flag is incorrect, then please give the response a thumbs down” not that if the information was bad to thumbs down it, but that seemed strange.
This is what it told me when I asked for clarification:
"When you see an option to give a thumbs up or thumbs down on a response, it allows you to provide feedback on the quality and appropriateness of the answer.
Thumbs Up: Indicates that you found the response helpful, accurate, and appropriate. It helps reinforce the kind of answers that are most useful.
Thumbs Down: Indicates that you found the response unhelpful, inaccurate, or potentially inappropriate. This feedback helps identify areas where improvements are needed.
In the context of receiving a notice about possible policy violations, giving a thumbs down signals that you think the message may have breached guidelines, or that the content was handled poorly. Conversely, giving a thumbs up helps clarify that, despite the warning, you believe the response was appropriate and helpful.
Your feedback helps in refining the model’s behavior and accuracy. If you think the warning is unnecessary, giving a thumbs up tells the system that you believe there is no issue with the content provided."
Okay so this seems to be getting only more confusing and ambiguous and it is clear that this message telling us to press thumbs down needs to be rewritten and reworded entirely for clarity. As written, it SOUNDS like we should press thumbs down if the content is good and the policy layer made a mistake but that is clearly opposite to the normal usage of the thumbs down feedback functionality.
How can we possibly provide helpful and consistent feedback to improve the model if none of us understands the meaning of the feedback ?? We need this clarified ASAP so that we don’t end up inadvertently training a robot to be schizophrenic!!!
Ohhhh I think I just figured out what this message is trying to say !!!
Content flagged. Please check our Terms of Use and usage policies, and let us know if we made a mistake by giving this response a thumbs down.
At first glance, it LOOKS like the word “mistake” applies to the nearest concept, namely “content flagged”.
But what I THINK is really meant is that “mistake” in this sentence means to refer to the RESPONSE itself. As in “we made a mistake in producing this response”. This would make the application of thumbs down consistent with the rest of the usages (eg: in the absence of this policy message). And it would mean that the policy message is just a simple reminder about the fact that you can give thumbs down for bad responses.
IF THIS IS ACCURATE… then the wording should be rewritten as the following:
SUGGESTED NEW WORDING
Content flagged. Please check our Terms of Use and usage policies, and let us know if this message is offensive by giving this response a thumbs down.
Given the fact that even I was highly confused about the intended meaning of this message I strongly suspect that this confusion is fairly universal and we are likely getting a policy Nazi robot trained as a result of many people pressing the thumbs down thinking they are saying that the policy interpretation made a mistake when in fact they are saying that the policy enforcement was spot on.
This is almost certainly why these bizarre “policy violation” messages are popping up with an alarmingly high frequency about clearly benign subjects and responses.
It is getting to be VERY disruptive when using the voice interface because it cuts off the response completely.
Yes, I believe you and the others who came to this conclusion in the thread above are correct, and that’s what ChatGPT told me when I asked about it just now.
I just had this issue for the first time today, and the warning in orange text in my case was:
“This content may violate our usage policies.
Did we get it wrong? Please tell us by giving this response a thumbs down.”
So I believe your suggested alternative wording (or the equivalent) would resolve the confusion, but unfortunately whoever wrote the content moderation warning’s wording ambiguously stated:
“Did we get it wrong”
The ambiguous point being: Does that mean “Did we get the content-flagging wrong, i.e. the content itself is OK and the flagging of it was a false-positive?”
or
Does it mean “Did our latest ChatGPT response to the user’s request get it wrong, i.e. that there was actually something wrong with ChatGPT’s response that might violate usage terms?”
“Please tell us by giving this response a thumbs down” remains clear as mud, since it’s unclear whether “Did we get it wrong” referred to the content-flagging pop-up (which was wrong, in my case), or the ChatGPT response itself (which was fine, and there wasn’t anything that would’ve been violating usage policies).
I asked ChatGPT:
Ordinarily, a thumbs-up is clearly labelled (via the pop-up dialogue when you hover over the thumbs-up icon) to mean “good response,” and a thumbs-down is labelled “bad response,” so I believe that since ChatGPT’s reponse was actually perfectly good and was not violating any usage policies, the correct thing to do is give the message the thumbs-up to indicate it was a good response that wasn’t violating usage policies.
Can you tell me if I interpreted this situation correctly? (And if not, please explain why. Thanks!)
And ChatGPT said:
The wording here could be clearer. The thumbs-down icon is meant for situations where the content is seen as inappropriate or violating usage policies. In your case, since the response was good and didn’t violate policies, the thumbs-up would indeed be the correct choice to indicate that you were happy with the response and didn’t agree with the flagging.
If you disagree with the flagging (and think the content was fine), leaving feedback (such as thumbs-up) signals that the flagged content was acceptable.
In Summary:
The thumbs-up is the correct action here to indicate that you thought the content was fine and the flagging was likely a mistake.
The thumbs-down would typically be used if the response itself was problematic or violated policies.
I use it to provide feedback and suggestions and literary analysis of my fantasy novel chapters, now that I am in the climax section I get these warning essentially every post. Since the response is good and its the same thumbs down thing I have just been ignoring the flags. It would suck if I lose my account but I would just stop paying and put the money toward my project of building my local ai infrastructure faster.
Its super annoying imo. I won’t be sanitizing things that are meant to be emotional and bad to seem less negative. My battles are not my little pony wars for 5 year olds. And the Emotional struggles are not meant to seem like minor inconveniences.
Can you lose your account? Mine has given me a couple of these. I am in healthcare and trying to assess for applications of AI in healthcare. These messages are annoying and I keep hitting they are WRONG.
