Hello,

i own quite a large community existing 25 years now with over 70.000 members and over 5.6 million posts and gigs of attachments,
In all this years, an incredible amount of knowledge was acumulated and i'm searching for a way to make this knowledge accessible.
I was thinking about training an AI with all my community data.

What do you guys think about this? Feasable?

Best,

Mike

Edited January 29Jan 29 by MaNiAc LRSC

While I don't doubt that there are many valueable posts in your forum I don't think the vast majority is. That said, you need to set clear restrictions and conditions for your model which posts to add or you don't get the result you want. If you do that it is likely from your 5.6 million posts only a few thousands are that valuable. If you don't do this you will get a lot of noise which makes your result below average.

Is there a possibility to create a feedback loop if the AI's answer is not correct?

I won't be abke to tag all valuable posta, but the users could by giving feedback.

Are there any AI engines existing suitable for my use case?

Unfortunately, how ML in general works I doubt this is easily possible. ML can do syntax but not semantic. In my experience LLM for a specific purpose where curated data was used to train them are mostly highly accurate. Also, they don't hallucinate an answer if there is no valid answer to your question. They just say, there is no information. What is also true is, usually, they're not free because the curating of data is done before creating the model and depending on the knowledge domain this could be a manual process which is expensive.

I can't help you with the question because I don't know your knowledge domain.

If you use IPS cloud service I think they're currently planning such a feature. For the same reason as stated above, though, I have my doubts it will work this well.

@MaNiAc LRSC - did you end up building something out?

We just had one of our Community Members ask if we could implement a LLM in our Forum so they could ask it questions, especially about helping find historical threads on particular topics.

No, lack of time.

But i think this would be a gamechanger.

@Mark How about Invision building such a feature?

 

No, lack of time.

But i think this would be a gamechanger.

@Mark How about Invision building such a feature?

You would need to post up a suggestion within the feedback section. Note, Mark hasnt worked for the company for some years now :)