OpenAl Plugin
-
We are focused on 3.x right now but we could build a plugin in the future that let's users ask questions to chatgpt by tagging it. For example
@chatgpt <insert your question>?And then it would reply to that post with an answer. Should be trivial to build using https://www.npmjs.com/package/chatgpt
-
That's the main thing that gives me pause... while I don't believe that ChatGPT will signify the end of online human interaction (among other things), I do believe that we will adapt to it and learn to live alongside.
However my prediction is that ChatGPT will rather have the opposite effect, that it will enhance the value of user-generated content, simply through sheer quantity of AI generated content. It doesn't make said UGC any more or less correct, but human interaction will always be more interesting than a human-like construct
Lots of people are doing very interesting things with ChatGPT, and I'll put it very very bluntly:
- Something we can do is have ChatGPT consume an entirety of a NodeBB's post content and answer queries like a tailored help bot
- Seems like we see a new library or script doing this almost daily
- It is easier for me to sit back while hundreds of people attempt to do this, and wait for the winner to surface
-
@baris this sounds very cool
tagging
@chatgpton a post is a better implementation that what I wrote here: https://community.nodebb.org/post/96921 -
@baris said in OpenAl Plugin:
Same post-queue rules would apply so not sure how it's different than some user creating an account and posting spam generated by chat-gpt
It stops being their API usage
But yeah, with the UX mentioned here it wouldn't really be abusable, but I feel it'd end up being a novelty rather than something useful.
My main issue is the question of when it'd be more useful to tag the chatbot than use its own interface.One idea is Q&A forums, where providing an "attempted answer" (especially if it's using data from the forum) and still having the opportunity to see the community clarify or correct that may be useful - but I'd say automatically feeding question topics for a response may be better.
Probably the most useful, but also more abusable, would be to integrate a chatbot with the composer, so that it's useful for making posts, rather than after one makes a post (in part also because my experience with even newer models is that they still need to be very highly guided to do anything even slightly complex, so the feedback loop needs to be faster than posts).
-
@cagatay I think NodeBB dev should not waste a single watt of their energy on this at present.
Discourse has gone all in with AI, the latest beta introduced their AI bot if you want it via plugins.
I say, let everyone else run off the cliff edge on this one.
None of this ChatGPT/AI stuff is at it seems, and it can not be judged or assessed in totality as some technical challenge - The vampire can only come into your house invited.
-
@omega said in OpenAl Plugin:
None of this ChatGPT/AI stuff is at it seems, and it can not be judged or assessed in totality as some technical challenge - The vampire can only come into your house invited.
I kind of agree - I hate that so many people seem to be treating "hallucinations" as some bug that can just be fixed and not part of the fundamental way these models work, and just cherry-pick the good outputs ignoring all the work that goes into getting them (seriously - how the hell do people program using ChatGPT? Like, Copilot is useful for boilerplate and just as a good autocomplete, but for actually writing anything more complex than algorithmic class assignments it takes me more work to get a decent code from a chatbot than just writing it myself).
But a good autocomplete is still worth something, and even the current "AI" can do a lot more. I consider them incredibly useful, but even more overrated, tools. This is why I'm suggesting using it not as a response generator (which will produce a lot of garbage, even if some results will be good and it's an easy way to say you followed a trend).
Also, another thing that both Discourse and NodeBB now have (see chatGPTmoderator plugin) is AI-based moderation, which is actually great - a lot of moderation is already somewhat automated, and this is just automation that works far better with text semantics and not just literal words or obvious usage patterns. It's not perfect by any stretch of imagination, but it can allow for more efficient moderation and that may be all that's needed.
-
@oplik0 Funnily enough (and by the way great post), Ai moderation for me presents one, if not the prefect example of the issues, the great dangers of the current mass embrace. I find it impossible to deal in this stuff narrowily because of the profundity of the implications, as written it is not a techicnal debate, it's about all of everything.
If you're Free speech, then AI and AI moderation has solved nothing for you. It has no benefit, and it has no place.
If you're not for Free Speech, then AI is a wonderful addition to exerting your power and it's gimme gimme gimme asap!
AI clarifies a lot of things, on all levels, and pushes us to the limits of very sense of existence and self and poses serious exmaintoin of WTF are we actually doing, what are we even here for, becasue, up to this point it's all been a human based impetus, building tech and the web for people, but what I see now is, people being impressed upon by perhaps becoming to close by spending too much time with their machines that their dreams are filled with vision of how to build a better world for robots, at all costs, even to people, because the people are what is broken and need to be fixed. Oh dear.
There is no middle ground IMHO, and the above example of freedom of speech indicates how it expresses in a derivative way. You are either all in on AI or you are not.
Then such calls for "regulation" from various CEO's is a pitch for middle ground. Beware those (Greeks) bearing gifts.
-
I mean, it's important to note that I'm talking about AI as moderation help, not sole moderation. Auto-flagging, or putting in moderation queue instead of making the post public.
And almost regardless of ideology, I hope most people see the need for moderation - at least on the basic spam level. And for most people this quickly escalates to illegal content, and some general content policies (things like obvious harassment, often nudity, etc.). And as the issue of the content that needs moderation gets worse (gotta love AI companies selling the cure for the disease they are themselves creating
) there is a need for new tools in that area.Ultimately, everyone wants to make a "free speech platform" but the way internet works is utterly incompatible with fully embracing this idea. And sadly, making it work differently may be much worse (and more dystopian) than just moderating.
-
@julian said in OpenAl Plugin:
@oplik0 it sounds like you're writing a response to a question about not being able to log in, would you like me to prefill a response about checking
config.json?@julian said in OpenAl Plugin:
Something we can do is have ChatGPT consume an entirety of a NodeBB's post content and answer queries like a tailored help bot
I was thinking something more like this... (sorry, since our forum is about US immigration, I will give example accordingly)
Q: How long can I stay out of US after getting Green Card?
A: General recommendation is not staying outside of the US more than 6 months since this will affect naturalization process. You can find more information on this topic here: www.forum.com/topic/1/...
and we can easily create a list of 200 items of FAQs...
-
@oplik0 said in OpenAl Plugin:
I mean, it's important to note that I'm talking about AI as moderation help, not sole moderation. Auto-flagging, or putting in moderation queue instead of making the post public.
Yea this was clear, and can you rule out moderation need and capacity (what is moderation but a edit-post-fact feature) but I suppose the way I frame it is in pre-gpt vs post-gpt world.
All those previously develoepd features or features in a pre-gpt mindset aren't counted.
@oplik0 said in OpenAl Plugin:
And as the issue of the content that needs moderation gets worse (gotta love AI companies selling the cure for the disease they are themselves creating ) there is a need for new tools in that area.
I did note that in the previous post and I'm glad I'm not the only one calling this out, it all seems a bit rich. Help we have no self-control, we need to be control (dominated). Yes, trust such I do not.
-
@crazycells There are many things to think about, including that which you invite into your house, but how will your own personal "wouldn't it be cool" GPT use-case for your web offering be better than this in the mid to long run? https://hu.ma.ne
