Should I be reporting AI-generated nonsense responses?

[TDK]

In instances where they don't answer the question, or where they answer it incorrectly, I have been reporting these. Should I keep doing so?

Does the ST community have a stance on whether or not AI-generated reponses are allowed? Is that stance in the guidelines and if not, can we put it there?

For example, clearly this is ChatGPT generated nonsense:

Another one, same thread even.

At first glance it seems like these answer the question, but they fall apart under inspection. These two are obvious, some are less so.

[waclawek.jan]

> Should I keep doing so?

I do, but, incidentally, in reporting exactly one of the above posts, I asked the same question (even worded similarly) together with the report...

Now policies on this site are decided by what appears to be part of ST's marketing, who then don't make their decisions based on strictly technical considerations and have followed "modern trends" in the past (see "modern feel and look" etc.) While those posts here are quite clearly just attempts at spam sanitized by the "first poster can't post links" (which sometimes appears not to work for reasons... ) policy, thus even marketing will dismiss them; a non-technical person may be lured to think they may be useful, both by the fad, and the apparent "knowledgeability" of these posts (underlined by the particular language it was taught to use, which again appeals to many nontechnical people - not coincidentally, it was designed to do so, these things are aimed to sell to masses primarily).

Regardless, my personal position is clear: by its principial working, there's no Intelligence in AI, thus it can't contribute wisdom. Where it appears to do so, there's already a substantial lack of basics, which I believe is not the case here. Would ST be willing to invest substantially in the basic, well-structured learning material accomplished by concise examples, rather than just pushing "libraries" and clicky generators, the difference between real knowledge and the glorified web search engine AI in fact is, would be crystal clear to anybody.

This is not to say AI can't be useful, just not in places where many people think now they may be - and they are deliberately made to believe so by the AI promoters.

As a particular example, AI is said to know to program, and is said to replace programmers in the near future (remember, AI needs to be sold, so needs to appeal to managers, who have the powers to buy). Those, who are scared by this perspective, should think 1. whether they do real programming with putting in structures and algorithms or just fill in forms; and then 2. whether they really want to work under management which can't recognize their value.

The fact, that AI appears to be able to lightning fast put together programs which do apparently impressive stuff tells much about the programming environment of today. It's dominated by tools which "does things" for the programmer, who does not need to and is not even willing to take responsibility for, so he's told he does not need to worry about documentation (which in turn does not even exist materially). This created a world, where doing outwardly impressive things requires excessive search among the vast amounts of libraries and functions, just because there does not exist a well structured documentation for them (as an example, see "doxygen" "autodocumentation" plague) nor a coordinated, comprehensive set of (again, concise and well-documented) examples. The mechanical search engine may appear to win in such environment.

Or, to put it into context of mcus, the value of programmer is not in the ability to type printf("Hello, world"); ,  but the ability to use that in the variable peripheral environment of mcus, the ability to use it in context e.g. while debugging, the ability to recognize where is it inappropriate to use (e.g. in memory-constrained applications, in interrupts), the ability to replace it with any other method as appropriate.

AI surely can type printf("Hello, world"); very quickly.

Have a nice weekend!

JW

 

[TDK]

> spam sanitized by the "first poster can't post links" (which sometimes appears not to work for reasons... ) policy

Had no idea that was a thing. I was wondering what the point of posting those answers was. I guess sneaking in a link makes sense, but then if that link was scrubbed it never showed up.

If the responses answered the question, I wouldn't care as much, but they're just non-answers. Like "ensure correct configuration," just generic advice with no specific meaning. Not unlike what you get sometimes when you call up customer support lines and they ask you to delete your cache, as if that will cause your device to start working again somehow.

> a non-technical person may be lured to think they may be useful, both by the fad, and the apparent "knowledgeability" of these posts

And that is ultimately why I think they are bad. They give the appearance of knowledge, yet none of the substance. It's just a facade.

 

 

> The fact, that AI appears to be able to lightning fast put together programs which do apparently impressive stuff tells much about the programming environment of today.

I haven't really seen this myself. There are certainly programmers who are only able to write code by finding an example and modifying it (and sometimes not even that) rather than writing anything themselves. So, I can see where AI-generated programs might help. Technology isn't there yet. Even if it were there, a lot of people can't explain what they want enough to have someone else write it anyway. Getting requirements is like pulling teeth in my experience.

 

Anyhow, thanks for the comments. Cheers.

[Peter BENSCH]

@TDK @waclawek.jan 

Thank you very much for reporting AI generated spam! We are now able to automatically block and remove a lot of URL posting spammers. But recognising AI spam interspersed in threads is still a challenge from time to time.

We are happy to receive reports from users of the respective threads, who should now also be informed by the community by email if this is allowed in their settings.

Thanks again and happy hunting spam!

Regards
/Peter

[TDK]

Thanks @Peter BENSCH, I'll keep reporting. Cheers.

[Peter BENSCH]

@TDK 

Forgot to mention: thanks for the second example of a spammer, which I was able to remove from the threads thanks to this friendly hint. If someone inadvertently replied to the spammer there, I will remove that reply from the thread as well so that future readers are not confused.

Regards
/Peter

[Lina_DABASINSKAITE]

Hi @TDK , 

Thanks for bringing this up & your efforts on reporting, it does not get unnoticed, and we take relevant actions. 

Does the ST community have a stance on whether or not AI-generated reponses are allowed? Is that stance in the guidelines and if not, can we put it there?

Yes, we are working on this and announce the guidelines soon.

[Tesla DeLorean]

Please flag all posts you see using Generative AI bots

Where did this post disappear too? Posted 4 days ago, and now disappeared down the memory hole. Google still finds it, but I had to really try..

https://community.st.com/t5/guidelines/why-having-generative-ai-writing-responses-on-the-st-community/ta-p/619353

[Lina_DABASINSKAITE]

Hello @Tesla DeLorean , 

After publishing the guidelines, we noticed that an additional clarification is necessary. We will put them back once done. Apologies for this confusion :)

[TDK]

I just reported a good 5-10 different AI responses that came in the past 10 hours or so.

I wonder if requiring a manual approval for a user's first post would be beneficial, along with a note to such users. Many other forums have this policy.

[Tesla DeLorean]

I report any that I'm sure are bots. Some disappear more quickly than others, and I assume Peter is generally on top of alerts whenever they come in.

I'd like to hope that 3 or 4 users flagging "inappropriate" posts would be sufficient to down vote them out of general visibility, but again this assumes a functional forum implementation.

The Policy thread was pretty good, and read only, so I'm not sure why it disappeared, and I was left with the Mandela Effect of having seen it but unable to find it again to prove I'm not insane..

[waclawek.jan]

Was there an AI response to this thread, too?

and it's "access denied":now.

That's hillarious, and I'm genuinely sorry I couldn't see it.

JW

 

[Peter BENSCH]

@waclawek.jan Yes, it was auto-deleted and so access to it is no longer possible. That's the only slightly confusing thing, that you get an email about a reply to a thread you were active in and then the reply in question is already in nirvana.

Even more unpleasant is when someone has already replied to it in between and only the spam is deleted.

[TDK]

I reported this one but it didn't get deleted, and then it got accepted as the solution. Great. Long term, if that happens more and more frequently, I don't know if I'll want to participate.

 

[Peter BENSCH]

@TDK Sorry that it took a little longer: the post has already gone to the eternal hunting grounds. I kindly ask you to forgive us for the delay (due to weekends, holidays, ...) and continue to report such findings.

Kind regards
/Peter

[Laurids_PETERSEN]

Hi everyone,

As @Lina_DABASINSKAITE mentioned, it was important for us that our terms and conditions were in line with our policy regarding generative AI.

The guidelines are now back online:
https://community.st.com/t5/community-guidelines/why-predominantly-ai-generated-posts-and-replies-on-the-st/ta-p/633785 

Thanks for your patience, and for your continued efforts to report AI-generated posts and replies.

Best regards,
Laurids 

[Andrew Neil]

@Laurids_PETERSEN - perhaps need a new option in the reporting 'Reason' drop-down?

 

[Tesla DeLorean]

I'd like to see it just be called "Report Content / Material"

And an option along the lines of "Review by Moderator" for stuff that's not Inappropriate, but needs someone to get on top of quickly, or assigned to appropriate staff.

There's a limit to what we can triage from the outside, but there are things that have enough gravity or recurrence that someone on staff needs to jump on.

I can click on "Something else" if we all understand that carries sufficient weight, and doesn't black mark the respondents in whatever secret metrics the back-end is reporting internally.