ChatGPT Integration tools for Discourse

[samp_one]

Dear Community,

As our discourse communities are mostly based on text, are there tools available to leverage AI like ChatGPT to get structured responses from discussions in communities?

I guess there are some integrations available for slack and discord. I think many admins would be willing to pay for such a bot. Sometimes intelligence gets lost in conversations. Like some AI tools summarize online teams or zoom meetings it would be great if we could summarize detailed conversations and let users talk to ChatGPT to get their queries answered.

Basically a generative search powered by GPT with the ability to summarize conversations would compliment discourse and add more value to users.

Just a thought. What do you think? Is it possible to implement anytime soon? I know I am asking for too much and being greedy. But what to do…

Take 6 Minutes to Create a free OpenAI GPT-3 bot via a Slack with Serverless...

[Falco]

As said elsewhere:

@Falco

For the next release cycle, we at Discourse have elected AI as of our priorities for new features and plugins. That means that we now have a dedicated team working on features like LLM integration, and will present those soon here on Meta.

You may also want to read

Integrating GPT3-like bots?

People, I see there are a few interesting topics on integrating bots at various levels into Discourse, but I did not see any topics about integrating the big, large-memory model AI bots like GPT-3, BlenderBot etc. Has anyone thought about this? or even made any progress implementing the idea? I am thinking of setting up another instance of Discourse with one of the big AI bots as an ordinary user on the system who participates like any other human user - but perhaps only replies when a post i…

Summarising topics with an LLM (GPT, BERT, ...)?

Dear discourse team, Discourse has the ability to summarise long threads. My understanding is that it uses engagement metrics like reactions or number of replies to identify what posts should be kept. In my community (internal communication within the company) this almost never triggers despite threads that are 20 pages long (reply count is not the best trigger for us) A more powerful approach would be to use an LLM like GPT-3 or BERT to summarise the whole thread on demand (vs just filtering …

[philip_rhoades]

This is excellent news for our Phi Rho Project:

http://forum.phirho.org

(PM me if you want an invite!).

I would also like to participate in the development work for Discourse itself if possible - I could be useful I think - depending what needs doing . . who do I speak to about that?

[twofoursixeight]

This could be useful. However, I have heard that ChatGPT can generate incorrect answers and potentially dangerous answers. I have also heard that it states wrong answers with equal confidence to correct answers. Also, what everyone else said. ChatGPT is also incapable of browsing the Internet. On the bright side, it makes answering questions a breeze as long as ChatGPT is fetching correct answers, and the question is clear.

[MarcP]

There is more then just answers.

For example, GPT could automatically handle flagged posts, and even take into account previous posts.

It can correct spelling errors and typos.

It can help with title suggestions/related topic suggestions.

Something useful that comes to mind is a feature to fact/double-check a post with GPT. Some button with a predefined prompt “check if the information in this post is correct”.

That’s just a few examples that came into mind.

[merefield (Robert)]

@MarcP

For example, GPT could automatically handle flagged posts, and even take into account previous posts.

That’s interesting, can you elaborate how this might work?

@MarcP

Something useful that comes to mind is a feature to fact/double-check a post with GPT. Some button with a predefined prompt “check if the information in this post is correct”.

Can you point to an existing service/API that does this?

Why not just use community flags?

This is what was proposed for Topic Summaries and I’ve implemented a bespoke downvoting mechanic (because the summary is distinct from a post)

How else to process without involving humans?

[Canapin (Coin-coin le Canapin)]

Hi there @samp_one

I think this particular link posted by Falco addresses your main question:

Summarising topics with an LLM (GPT, BERT, ...)?

Dear discourse team, Discourse has the ability to summarise long threads. My understanding is that it uses engagement metrics like reactions or number of replies to identify what posts should be kept. In my community (internal communication within the company) this almost never triggers despite threads that are 20 pages long (reply count is not the best trigger for us) A more powerful approach would be to use an LLM like GPT-3 or BERT to summarise the whole thread on demand (vs just filtering …

I’ll set this support topic to be closed, but feel free to tell us if you think the linked discussion doesn’t provide the proper answers or discussion you look for