Re: What to do, if chatgpt gives a wrong answer

With all respect, this seems very much like solving a problem that doesn’t exist.

This community leans (heavily) towards the technical, and in my experience that’s the crowd that’s least likely to use tools like ChatGPT to answer technical questions.  I’m not saying it’s completely useless, but anyone using it to identify command flags is probably doing things “incorrectly” already.

Better to spend time fixing errors in the actual product/documentation than trying to train a model we ultimately have very limited (at best) control over.

The only way to use ChatGPT correctly is to distrust it entirely.  It’s handy for quick reminders regarding information you can already vet, but no, please don’t use it as a replacement for the actual docs, the community, or the actual underlying code.

That’s my two cents.

- C


> On Oct 11, 2025, at 4:14 AM, ft <ml@ft-c.de> wrote:
> 
> Hello, 
> 
> When I ask ChatGPT a question about FreeBSD, I very often get the wrong
> answer. Does FreeBSD have a plan for improving the quality of answers
> so that it can be taken into account in the next version?
> 
> Here are some suggestions (which I received from ChatGPT) on how this
> could be done:
> 
> First, use the built-in feedback system. If you are using ChatGPT in a
> browser, you will usually see thumbs-up and thumbs-down buttons under
> each answer. If an answer is incorrect, click the thumbs-down button.
> In the feedback form, explain what is wrong and provide the correct
> information. For example, if ChatGPT suggests an invalid option for the
> pkg command, you could write: “The response about pkg usage is
> incorrect. The option -b does not exist as described. The correct
> option to build a package is -p, as documented in man pkg(8).”
> 
> Second, it would be great if the FreeBSD community could create a list
> to collect and document wrong answers. Keeping a simple personal record
> of mistakes is also very helpful. The idea is to systematically
> document any wrong answers before submitting them as feedback or
> sharing them with others. This makes it easier to prove what was
> incorrect, ensures you don’t forget important corrections, and helps
> others understand the issue clearly.
> 
> A practical way to do this is to note:
> - The topic, command, or function the answer relates to, e.g., “pkg
> build”
> - The incorrect answer given by ChatGPT, e.g., “Option -b builds a
> package”
> - The correct answer, e.g., “Option -p builds a package”
> - The source or reference, e.g., “man pkg(8)”
> - You can collect this information in a simple text document or table.
> Later, when giving feedback to OpenAI, you have all the information
> ready and verifiable.
> 
> Third, use community channels to amplify the feedback. OpenAI’s
> community forum (community.openai.com) allows you to create threads
> about repeated errors or FreeBSD-specific issues. Other users can
> comment, verify, or add additional information. For open-source AI
> models, you can submit issues directly on GitHub or the project’s issue
> tracker.
> 
> Finally, when giving feedback about FreeBSD, try to provide context.
> Include the exact FreeBSD version, quote commands or man pages, and
> explain why the original answer is incorrect. Logs or screenshots can
> also be helpful.
> 
> By following these steps, we can gradually improve the quality of AI
> responses for FreeBSD topics and help future users get more accurate
> information.
> 
> Best regards,
> Franz
>