7
$\begingroup$

In the last weeks, I have seen about half a dozen answers that do not answer the questions properly and just read strange. I ran them through writer's AI content checker and for the worst ones got scores of 19% to 45% Human generated text. Copyleaks AI checker scored the same texts that the likelihood of a human origin was less than 80%, in one case even showing up as only 19%.

The problem is, that the solutions offered in those texts are at times totally nonsensical. This has to stop.

I propose to formally ban all AI-generated answers for the time being and to delete AI generated content as it is detected.

$\endgroup$
0

3 Answers 3

Reset to default
4
$\begingroup$

As a community we decide what we allow or don't allow in our stack. Voting is the means by which we do this, so, if maleficent or AI generated content is being generated, just vote down and vote to close (as it doesn't answer the question). The system will take care of such users or mods will pick this up to process further.

Fact is that we need to be aware that not everybody is as fluent in English, so, we need to take care to misjudge posters on a single answer or question. But, generally, good point to be aware of.

$\endgroup$
1
  • $\begingroup$ AI content has patterns and is remarkably different than clumsy english human content. On SO, AI generated content is (temporarily) banned and that's a good thing. $\endgroup$
    – Soleil
    Commented Mar 24 at 16:14
2
$\begingroup$

TL;DR - I agree with the proposal

Note that I found it was rather difficult to formulate an answer to this question, without making it sound negative, opinionated and panic inducing. In fact I decided to remove quite a bit, and just leave in factual links and quotes.

I came across this meta question over on SO today, How can we determine whether an answer used ChatGPT?, and this answer in particular grabbed my attention:

I don't think there's going to be any value in having you, a mere mortal, try to determine if an answer or question was generated by ChatGPT.

The reason is super, super simple: we don't want this to turn into a witch hunt.

There are things you can do even today to help with curation of answers; if an answer is wrong or incomplete, you can downvote it. If you happen to notice a trend of similarly bad answers from an account, you can flag for moderator attention with the series of answers you've got as evidence.

This shouldn't be a "those bad answers are more moderatable than others" situation - just vote on answers as you always would, and let the moderators deal with the heavier lifting on this circumstance.

Which raises a valid point of allowing suspicion and doubt to get the better of us.

Yes, we have recently seen some questionable posts, and (to be frank) even now I am still not 100% convinced what the root cause of those posts were.

Nevertheless, community voting and flagging seem to allow us to handle such situations reasonably well.

However, one thing that is certain: ChatGPT, and any other derivatives, are not going to go away, and it is probably going to become more and more difficult, as time goes by and the technology "improves", to detect such posts.

The tools that you used (and refer to in your question), while I haven't tried them myself, look useful - even though they themselves would also appear to be using some sort of artificial decision making! The best tools are probably our own eyes and knowledge to root out suspect sentences and phrasing as well as (more importantly) suspect advice.

In addition, a couple of answers (to the linked to meta SO question) also point out some good tips and what to look for - such as this answer and the first paragraph of this answer, in particular.

$\endgroup$
0
$\begingroup$

Can you show examples of this? I wasn't aware and it sounds baffling.

Are there robots crawling the site and posting garbage answers just because? Or are they posting spammy links for example, like the good old days when we had blogs and the comments would be full of those unless we enabled a captcha?

Also, being an ESL myself, 0scar made a good point. Sometimes people have wonky English but they are still human.

$\endgroup$
1
  • 1
    $\begingroup$ most of the content has been worked on by now so that neither the AI detection triggers nor does it read like garbled nonsense that misses the point of the question anymore. Think about a print volume to quality spreadsheet for an SLS printer while discussing an FDM printer and talking about temperature in the extruder on a bowden style machine (which is the motion system of the filament. that is only close to the meltzone in a direct drive machine) - Things that triggered mental images of fleeting knowledge about some aspects but missing blatantly obvious other parts. $\endgroup$
    – Trish
    Commented May 21, 2023 at 8:02

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .