ChatGPT in forums

[Matt C.]

I'm posting this thread because of two add-ons released on XenForo today.

[021] ChatGPT Spam Filter

I am excited to offer you a solution to help you combat spam on your forum. This add-on is designed to seamlessly integrate with XenForo's standard spam filter system and utilizes the ChatGPT API to quickly and accurately filter out spam messages...

xenforo.com

[021] ChatGPT Autoresponder

Extended version for communication: AI Bots Designed for ✅ Quick responses from ChatGPT to threads in specific nodes ✅ Clarify answer details with quotes Not designed for ❌ Chat with the bot on the forum ❌ Reply to mentions I recommend paying...

xenforo.com

The first one seems less harmful. It filters messages and decides what's spam using ChatGPT. The second one seems more dangerous. It can reply to first post in a thread.

The one I really want to discuss is the second one. One hand, the technology is amazing and has many use cases (not including forums). but the other hand, it has dangerous effects. It is now starting to make its way into forums. I can see a couple legitimate use cases. If you have a question/answer forum, ChatGPT can reply with the answer and the community can upvote/downvote the answer. It should be known that the answer is from ChatGPT.

But sooner, rather than later, I guaruntee you someone is going to make an add-on that fully mimics users and they have full on conversations. That part is scary, because the dialogue may get so good, that you won't be able to see what's real and what's not. Admins will be able to generate fake activity much more easily.

What's everyone opinion?

 

[Tracy]

If I want to talk to a bot, I'll do it directly. Forums already have an issue with actual participation. Why in the HELL would ANYONE want to do such as replying to threads. I'm sorry, even an AI doesn't have the REAL WORLD experience that most knowledgeable repliers would have. Exactly how many times has that AI taken out an ASI533MC Pro or an ASI533MM Pro and taken actual images with it with a specific scope?
We might as well wipe the internet and then refer ALL people to an AI chatbot for ALL sources of data.

I can see it, maybe, as a spam filter possibility.. but even that would be stretching it. There are already well enough resources that can be used for "common" spam type posts.
It's just the "latest fad" and everyone is wanting to jump aboard the train.

 

[Matt C.]

But sooner, rather than later, I guaruntee you someone is going to make an add-on that fully mimics users and they have full on conversations.

And...it's here.

[021] ChatGPT Reply Bot

This add-on uses ChatGPT to generate replies to posts automatically, providing instant and intelligent responses to keep the conversation flowing smoothly. With its advanced contextual analysis, this add-on ensures that the generated replies are...

xenforo.com

 

[Cedric]

Honestly, copy and paste as a person is not done. Can it serve purpose on expanding a subject? Yes. Can it serve as a purpose to provide a quick reply and being informative? Yes. Is it always correct? No. But us human are not always correct either. So you may not welcome the add on, that’s your right but don’t hate on the dev providing it. What purpose you use it for is entirely up to the user.

 

[Kale]

I think it is fine by me as said above. I honestly would be open to trying it, but would plan to be open and honest about this add-on and do it in some forums/sections.

 

[Arantor]

I don't think it's a good idea to encourage it at all. Honestly...

1) If you're doing a support site and it's an automated response to a question, great - if it answers the question. It almost certainly won't, because it probably won't understand properly, but I suppose it might cut down a few % of questions.
2) More importantly, the AI generated blurb will subsequently end up being fed into the next iteration of these things - it is already happening, but not in such a huge quantity that it's (currently) a problem. But it will become one where the infinite regurgitation machine is being fed on previous regurgitation.
3) For things not 'support' where you want to have an actual conversation and discuss actual opinions, AI simply doesn't understand - it's not that it doesn't understand enough, it has no comprehension whatsoever, it is just throwing things at the wall and seeing what matches up statistically to feed back to you. This is not a conversation I want to have, if I want to talk to a machine that doesn't understand, I'll fire up an Eliza from the 1960s and have about as much fun.

For giggles I asked it tell me about me. What I got back was... at best half right. It did big up my credentials somewhat, suggesting that I was the lead developer for SMF (hahahaha) for many years (nope) and that I was a significant contributor to SMF 2.0 (I was, but not nearly as significant as I was to 2.1). It grudgingly conceded I was right when I corrected it, but very grudgingly so and if it were human it would be that pouting-face-kicking-the-can 'oh ALL right' vibe. But it's not, it's just repeating words it's heard before.

 

[Oblivion Knight]

It isn't the first time that such bots have existed, I distinctly remember one or two being available for vBulletin back in the day (2.x or 3.x)..

It's up to individual admins whether they wish to utilise such functionality. I really don't mind either way - and some of those complaining likely have paid posters who make contributions of a similar quality, if not worse.

 

[Nomad]

I think a spam filter bot will be more useful for forum owners that the bot that replies to the first message. If the bot replies, what is the use of a forum where humans take pleasure in interacting?

 

[Arantor]

I can see a use in support forums, I guess, trying to answer the question, trying to direct the user to the right place - and it's better than any of the bots from back in the day, for sure. But support/Q&A forums don't really react like most other forums - the majority of people come, ask their question, get an answer, bounce off again.

 

[Tracy]

I can see a use in support forums, I guess, trying to answer the question, trying to direct the user to the right place - and it's better than any of the bots from back in the day, for sure. But support/Q&A forums don't really react like most other forums - the majority of people come, ask their question, get an answer, bounce off again.

The biggest issue.. there is the proven lack of providence of the answers provided by the AI. With REAL human interaction, you get a wide variety of points... which in my niche are ALL valid.
Captures in a Bortle 1 area are NOT the same as in a Bortle 9. So a generic answer is basically useless.

 

[Arantor]

Yup, and as someone who's done support, I'm not in any way convinced either, but if you imagine a less technical field where the answer doesn't have to be perfect but as a routing function to get people to the right place? That could have value.

 

[Matt C.]

The biggest issue.. there is he proven lack of providence of the answers provided by the AI. With REAL human interaction, you get a wide variety of points... which in my nice are ALL valid.
Captures in a Bortle 1 area are NOT the same as in a Bortle 9. So a generic answer is basically useless.

I bought it out of curiosity. Based on this, it has a long way to go.

 

[Kyng]

I feel like it's a bit of a double-edged sword.

I think a good way to use ChatGPT would be to generate the first draft of your post with it, then go through and edit it. That way, it'll be possible to generate quality content in a fraction of the time it takes to type it out from scratch. But if people are going to skip the editing process altogether, and just post unedited ChatGPT output... then honestly, it'll probably just come across as spam.

There's a wide range of potential outcomes here, from "This will vastly improve the forum experience" to "This will turn everyone into a spambot, thus completely  ruining the forum experience for me". I expect the eventual outcome will be something in between - but right now, it's a case of "Wait and see".

 

[Kale]

I feel like it's a bit of a double-edged sword.

It is going to be a very double edged sword IMO too. I think if you did what you said, then sure no problem. I think if people are going to use it to make a first post or so, then yeah some may class it as spam.

 

[Matt C.]

Well, looks like the Apocalypse has arrived.

[021] AI Bots – ChatGPT, DALL-E 2

Artificial intelligence on the forum? Ooh yeah :cool: A powerful tool that will blow your forum activity to the skies 🌠 Thanks to two built-in handlers: ChatGPT and GPT Dall-E. Bots on these handlers are capable of: Reply in threads to quotes...

xenforo.com

I may purchase to try it out.

 

[Cedric]

To be fair, I’ve been tempted to purchase as well. It’s pretty featured packed. And while not everyone may agree, if setup properly it can be a major asset to a forum.

 

[fords8]

Well, looks like the Apocalypse has arrived.

[021] AI Bots – ChatGPT, DALL-E 2

Artificial intelligence on the forum? Ooh yeah :cool: A powerful tool that will blow your forum activity to the skies 🌠 Thanks to two built-in handlers: ChatGPT and GPT Dall-E. Bots on these handlers are capable of: Reply in threads to quotes...

xenforo.com

I may purchase to try it out.

To be fair, I’ve been tempted to purchase as well. It’s pretty featured packed. And while not everyone may agree, if setup properly it can be a major asset to a forum.

I am interested to see this at work. I am not sure if I would buy it.

 

[Ravenfreak]

You know honestly I think having these plugins installed on a support forum wouldn't be too bad of an idea. Especially if the forum is very active and staff can't assist their members right away. Also installing the AI Bots plugin on a roleplaying forum that is brand new can help jump start the community if the admin can't find other people to participate right away.

 

[Arantor]

Support? The only context I'd ever want AI bots is to direct people to existing content; it's more likely to give them bad advice confidently than it is to give them good advice. (Not that it can't give good advice, because it can, but it doesn't have any understanding, it'll just be looking for words that match.)

Roleplaying... I might try that in a private board when my next site gets set up. I rather suspect it'll die on not having enough material to work with and leaving the other person not even remotely enough material to do anything with. But we shall see. (I also think it'll die on not having access to any of the lore.)

 

[Arantor]

So I had a go with this from an RP perspective... while I think it's not something I'd *personally* want to engage with, I think it's interesting enough that some people might.

To clarify, I gave it a paragraph of info about a character, with some prompting around the character's style and personality and abilities, and gave it a post to reply to. What I got back was very interesting - it made some very interesting jumps of personality on the part of the other person (so it's completely no go for most RP scenarios since it is absolutely happy to god-mode, very much a no-no) but it did make some very interesting inferences based on what I *didn't* tell it. Specifically it figured out the profession of the other character and adopted much of the tone one would expect from that profession, but this is hardly surprising from a statistical analysis perspective - I just found it *interesting*.

I'd share but to provide sufficient context would be quite detailed and lengthy. But it's not the disaster I expected - but it also needs some safety rails for sure, and it needs someone to build a character around giving it enough prompts to do anything useful.