Do you tolerate AI content on your forum?

[Deleted member 184953]

The problem with forums (at least many if not most of them) has been traditionally that they do have some gems of information, but they are covered with tons of mud and false information. Many forums are a gigantic echo chamber, full of beliefs, wrong factual statements and endless rumors

I thought you talk about Facebook or X

 

[StarArmy]

If the content is useful, it's fine. I judge it by the same standards as any other post. It's got to contribute meaningfully.

I'm even offering Reply with ChatGPT as a feature for user upgrade subscribers.

On my forum it seems mostly used by a handful of members with spelling and grammar issues as a way to pretty their posts into more readable posts.

The only issue I've had is that AI content tends to have a weird tone to it that's not really fitting with the casual vibe of the forum and stands out.

 

[FTL]

This is to lose all the interest of the forums, an AI is rarely wrong and therefore the answer to a question asked has a good chance of getting everyone to agree. And so the thread stops, in most cases. This is no longer an answer but a search result... On an active forum even if you ask the question how much is 2+2 then you can have a 3 page long thread! AI, if we let it happen on the forums then they no longer have any reason to exist.

Hmmm... that's a good point. It's like a radio talk program choosing controversial topics to discuss to keep the discussion going. I still think it's a grey area where the details matters. For example, I manage a cat forum (I'm not the owner) so if someone wants an answer to a cat care question, especially when it comes to healthcare and nutrition, then I want the most accurate answers possible so that the cats don't suffer from bad info.

On the other hand, if it's a discussion about Elon Musk's Starlink satellite internet access system, then a wide range of opinions is great in keeping the conversation flowing. I also don't enforce staying on topic too much as that can stifle conversation. Chat drifts in face to face conversations, so I think it's reasonable to let it do that on a forum, within broad limits, of course.

We didn’t find this to be the case when we CG tried the ChatGPT addon. Had it got more answers correct we may have continued with the addon but no.

It's true about current AI, and the companies behind the various beta AIs are keen to point this out so people don't rely on them too much. However, AI will only become more accurate in time. I reckon one day it could become the reference standard. When, I have no idea.

 

[Sim]

We had a situation where someone was using AI generated content to answer tax related questions on PropertyChat. We had a tax lawyer complain that the information was factually incorrect and pointed out that this type of thing leaves us open to legal issues.

The problem is not so much that the information was incorrect - people post factually incorrect stuff online all the time - the problem was that it was posted as if it was an authoritative source of information - and was thus misleading.

The whole question of giving financial advice online is fraught with issues at the best of times - throw in questionable generated AI content and it simply compounds the problem.

We now have strict rules against posting AI generated content on the site - and our members are pretty quick to pick up anything which smells generated.

 

[Sim]

Independent from forums I did a test with chatGPT already a while ago with facts about a topic that I am knowlegable about. Facts that are a bit complex to research and aggregate but there is a clear true and a clear false. It turned out chatGPT got the right answer within seconds - but only in about 60% of the cases. In the other 40% it claimed utter bullsh*t - clearly false facts and numbers - but only reconizable if you already are an expert in the topic it was talking about. So it is really dangerous as many people beliefe what the AI says.

This is the real danger - generative AI systems will happily provide factually incorrect information with complete confidence and written in an authoritative manner - to the point where it takes a real expert to identify that the information is inaccurate.

This can be extremely dangerous if people start relying on this information to make important decisions when they have no way of detecting whether the information they have been given is actually reliable.

 

[GW2]

...We tried the AI add-on, but soon retired it. Too many wrong answers and I think this can “kill” the forum eventually. We don't need AI on forums, we need human interactions, discussions etc.

Exactly. This is why I do not like Facebook. Not that Facebook is AI, but Facebook allows posting of almost anything, anywhere whether factual, misleading, or false.

That is why my forum continues to hold members' interest and draw new members. My forum is organized by subject matter with open, human interactions, discussions, agreement, and disagreement. And it is all human. My forum does not need AI, and AI could not substitute for the up-to-date experiences that my forum owners post daily.

 

[Deleted member 184953]

Something that might be interesting is to create a custom field, just a checkbox: Does your message include AI-generated content?

Obviously some will not check it even if this is the case but perhaps the majority will be honest.

And why not create a BBcode like for quotes but to publish content from an AI.

 

[Seeker-Smith]

Reply with ChatGPT

Is there an addon or how hard is it to add?

 

[StarArmy]

I got it from the Ukrainian developer 021 but they had some sort of beef with the Xenforo staff and stopped developing for XF as far as I know, and they also fled Ukraine to avoid military service so I'm not sure what's going on with the add-on's status or future now. It's surprising someone else hasn't developed an alternative.

 

[Deleted member 184953]

Is there an addon or how hard is it to add?

Ask to ChatGPT

 

[Seeker-Smith]

I got it from the Ukrainian developer 021 but they had some sort of beef with the Xenforo staff and stopped developing for XF as far as I know, and they also fled Ukraine to avoid military service so I'm not sure what's going on with the add-on's status or future now. It's surprising someone else hasn't developed an alternative.

Thanks for the info.

 

[Forsaken]

This is to lose all the interest of the forums, an AI is rarely wrong and therefore the answer to a question asked has a good chance of getting everyone to agree. And so the thread stops, in most cases. This is no longer an answer but a search result... On an active forum even if you ask the question how much is 2+2 then you can have a 3 page long thread! AI, if we let it happen on the forums then they no longer have any reason to exist.

ChatGPT (and every other AI model) are currently only slightly better than chat bots that existed with AOL and MSN Messenger. They have a lot more data, but they are still extremely prone to hallucination and being fed incorrect data. A lot of the way that these models are being built also show very inconsistent or incorrect ways of processing data such as the "how many R's are in the word strawberry" example.

Early LLM models that came out within the last few years were much more curated before the arms race started, as showing off accuracy was more important than being first to show greater success, which is now the focus of every company.

The only true use I find for current "AI" would be for generating a boiler plate, which will then be checked and refined (basically taking the role of an intern), or data processing.

 

[ForumFan]

I haven't had any, yet. But if I do... ban hammer. I have no use for that.

IMO, it's quite easy to spot.

 

[gizmo4321]

In the other 40% it claimed utter bullsh*t - clearly false facts and numbers - but only reconizable if you already are an expert in the topic it was talking about. So it is really dangerous as many people beliefe what the AI says.

This is my problem:
I've played around with using ChatGPT to generate code for various (relatively trivial) applications. WHEN it generates code that actually does something other than just spit out an arbitrary result, it is almost always wrong, and in ways that are not immediately obvious. I spend more time, or at least as much time, checking the code for correctness as I would have spent actually writing the code myself.

Admittedly I've been writing software for 45 years, both professionally and as a hobbyist, so I'm sure some of this comes across as "Hey you darn kids! Get off my lawn!". That being said, I'm greatly concerned by the use of AI for anything at this stage, simply because I can't trust the output.

AI is in general producing utter crap, but people are using the crap as if it is accurate, and by the time they figure out it isn't, they've crashed a network, or compromised security (I've personally seen both happen just this past year).

People don't need help from AI to make disastrous mistakes; we're good enough at that on our own.

 

[frm]

I've played around with using ChatGPT to generate code for various (relatively trivial) applications. WHEN it generates code that actually does something other than just spit out an arbitrary result, it is almost always wrong, and in ways that are not immediately obvious. I spend more time, or at least as much time, checking the code for correctness as I would have spent actually writing the code myself.

I'm new to PHP (well, I was a master at PHP3, but that all went out the window when I took a break for another career path). I recently got back into writing PHP, and, back to old habits of how I would have structured and formatted the code.

I learned from PHP.net what I needed to do to get something done, and it worked from the garbage I put together. However, I wanted to see where ChatGPT could help. I asked it to optimize my code for PHP 8.3 and it reduced lines by like 1/2.

It then made me wonder about the execution time of the script. So, I added an output for that for my old code and new code. The code I learned and wrote from PHP.net took 0.2-0.3 seconds (a small application, but a heavy loop, and files to write to) to 0.02-0.07 seconds.

I took note of the functions ChatGPT taught me and put those in my knowledge bank, as to not repeat the same mistakes, or to know what to do differently next time.

It's surely going to change coding...

I wondered if you put the entire Xenforo script through ChatGPT if it would be even more efficient. But, I don't think it would do much there, to be Frank. As a gimmick, I put /XF/AdminSearch/Searcher.php (2.2.8PL2) through the same way and it took it down from 169 to 126 lines. To be fair, it did strip the 14 lines of comments. Also, I'm unsure if it would still function if I replaced the file (and knew where to go to see if it worked the same). So, it could improve XF... if the devs used it and didn't rely solely on their learned knowledge. Though, it could become a crutch, so it's a double-edged sword.

Edit: I don't know if I'm allowed to post the main file and the ChatGPT revision due to copyright infringement to see if any coders could see if it would actually work.

 

[Seeker-Smith]

For my sites purpose its strictly entertainment.

 

[Forsaken]

I'm new to PHP (well, I was a master at PHP3, but that all went out the window when I took a break for another career path). I recently got back into writing PHP, and, back to old habits of how I would have structured and formatted the code.

I learned from PHP.net what I needed to do to get something done, and it worked from the garbage I put together. However, I wanted to see where ChatGPT could help. I asked it to optimize my code for PHP 8.3 and it reduced lines by like 1/2.

It then made me wonder about the execution time of the script. So, I added an output for that for my old code and new code. The code I learned and wrote from PHP.net took 0.2-0.3 seconds (a small application, but a heavy loop, and files to write to) to 0.02-0.07 seconds.

I took note of the functions ChatGPT taught me and put those in my knowledge bank, as to not repeat the same mistakes, or to know what to do differently next time.

It's surely going to change coding...

I wondered if you put the entire Xenforo script through ChatGPT if it would be even more efficient. But, I don't think it would do much there, to be Frank. As a gimmick, I put /XF/AdminSearch/Searcher.php (2.2.8PL2) through the same way and it took it down from 169 to 126 lines. To be fair, it did strip the 14 lines of comments. Also, I'm unsure if it would still function if I replaced the file (and knew where to go to see if it worked the same). So, it could improve XF... if the devs used it and didn't rely solely on their learned knowledge. Though, it could become a crutch, so it's a double-edged sword.

Edit: I don't know if I'm allowed to post the main file and the ChatGPT revision due to copyright infringement to see if any coders could see if it would actually work.

Of those 29 lines of code, how much was structural syntax changes? Also if you did not test if it worked, how can you know it is improvements to the code base?

Removing lines of code at the cost of legibility, cleanliness or documentation is not improving anything, it is just pointless optimization that generally does very little.

 

[frm]

Removing lines of code at the cost of legibility, cleanliness or documentation is not improving anything, it is just pointless optimization that generally does very little.

It's definitely not just removing lines, but I'm unsure if this small portion that I feel safe enough posting here would do the same as intended.

If so, it clearly made the code more efficient. However, would implementing the change breaks something somewhere else, and if so, would finding it and optimizing it fix the problem?

Original:

            $phraseFinder
                ->where('language_id', \XF::language()->getId())
                ->where('title', 'like', $groupPossibilities)
                ->where('Phrase.phrase_text', 'like', $phraseFinder->escapeLike($text, '%?%'));

            foreach ($phraseFinder->fetchRaw(['fetchOnly' => ['title']]) AS $match)
            {
                $title = $match['title'];

                list($group, $id) = explode('.', $title, 2);

                if (isset($phraseGroupContentMap[$group]))
                {
                    $contentType = $phraseGroupContentMap[$group];
                    $matchedPhraseIds[$contentType][] = $id;
                }
            }
        }

        return $matchedPhraseIds;

Modified:

            $phraseFinder
                ->where('language_id', \XF::language()->getId())
                ->where('title', 'like', $groupPossibilities)
                ->where('Phrase.phrase_text', 'like', $phraseFinder->escapeLike($text, '%?%'));

            foreach ($phraseFinder->fetchRaw(['fetchOnly' => ['title']]) as $match) {
                [$group, $id] = explode('.', $match['title'], 2);
                if (isset($phraseGroupContentMap[$group])) {
                    $matchedPhraseIds[$phraseGroupContentMap[$group]][] = $id;
                }
            }
        }
        return $matchedPhraseIds;
    }

I'm honestly not good enough at PHP, or smart enough, to understand whether taking these out of a function and putting them in brackets makes the code any much better:

 list($group, $id) = explode('.', $title, 2);

[$group, $id] = explode('.', $match['title'], 2);
Which stuck out to me the most.

 

[Forsaken]

Yeah, so most of the changes in those snippets are just syntax structure changes, with some code changes that would need to be tested.

But overall, there is no improvement in that code that would be any better than what exists, and even from a file size standpoint there really isn't much point as it lessens legibility (at least in so far as reading through large files of code).

 

[frm]

But overall, there is no improvement in that code that would be any better than what exists, and even from a file size standpoint there really isn't much point as it lessens legibility (at least in so far as reading through large files of code).

Would the difference between the two help with execution time though? I'm really not sure on that. If even minimal, could you hypothesize the amount of CPU saved for an entire forum? I'm going to take a guess and say it's not noticeable, but it might be on a big board.