Discourse AI has been out for some time now and I am trying to get an understanding of how people are using it. As the Product Manager for the AI team, I want to learn from your experience so we can make Discourse AI better suited for your community
If your community uses Discourse AI, I would love to hear from you!
I would like to know how your community is using all the different AI features available. At a high level I would like to know…
Understand the problems you are looking to solve by using Discourse AI
Which AI features are you using currently and why?
Which ones are working well for you? enjoying or disliking any of them?
What you would wish you change or see done differently?
Yes, let’s meet for a 30-minute interview via Google Meet
Yes, I’ll share my feedback in writing (PMs)
No thanks, not for me
0voters
I’ll be reaching out directly to anyone who in interested. Feel free to reach out to me as well, my inboxes are open!
Haven’t done much research in your AI offering yet, but the two main points which come up:
on demand translation to users language
answering simple questions by referencing to my custom answer corpus (the forum plus all documentations I provide)
on larger sites i could imagine a more strict/smart „already asked“ detection, however on my site I feel it’s better to have less posting barriers and rather have mods merge topics in a friendly and false-positive free way (maybe send mods such a merge suggestion?)
For the second point above, if you provide a infra to feed knowledge and answer questions, that would also be a neat „iFrame“ for other sites.
We do currently offer Discourse Translator
that has a plethora of providers and configurations
We are in the process of trying to build out AI Bot with RAG on uploaded documentation to help answer common questions, for now the Forum Helper persona on the AI Bot can do a decent job!
I wanted to note that I believe I have reached out to everyone thus far on the polls - will keep it going for the next coming weeks (If I haven’t or missed you feel free to DM me!)
Please forgive - I’m not familiar with the abilities of the Discourse AI yet. I have been wanting to consider this feature for content moderation, but have not yet looked deeply into its positioning or abilities.
We have an extensive list of prohibited language but AI moderation as a front-end to this would eliminate a ton of regex and keywords.
I don’t know if the Discourse AI can pre-process content, to trigger a function where we can determine what to do based on findings. Consider the moderation features in Discourse for blocking terms versus holding content for moderation.
I would like to get to a place where each of us can use the tools available to fine-tune how we manage these concepts:
What is acceptable content?
What happens when specific kinds of content are detected? (Trigger functions)
What API functions can we use to react to specific kinds of content triggers? Examples include: Warn user, or block user, or block based on number of warnings, or forward to moderator, or put in moderation queue, or combinations of these actions.
As a developer who is familiar with the current state of LLMs and related tooling, I lack two things related to all of this:
I lack a firm understanding of exactly where the Discourse AI fits, the specific model being used, and what coding/API options we have available for integration and management.
I lack an understanding of API options available that would allow someone like me to code fine-tuned solutions, rather than relying on the fine team here to consider and provide solutions that would hopefully be acceptable to the widest possible audience.
So “How do you use Discourse AI?” Well, this is how I’d like to use it. As to “make it even better”, I’m hoping that detailed documentation and an API will allow for site-level configurations that will eliminate a lot of team development here. That will allow the team to focus on things that we on the outside cannot do. I consider that a strong win-win for everyone.
The earlier bits about Discourse AI functionality/capabilities and the specific models used should be available within the documentation and the AI page
The latter comment about coding/API options to open up flexibility something we might look into doing in the future! I would agree with your sentiments about potentially being a win-win situation if and when we do get there
I noticed that if Gemini declines to summarize a post due to NSFW words contained in it, there is no error thrown, nor does it say anything. It just gets stuck loading forever.
I suggest some sort of a “Sorry I cannot summarize this content due to inappropriate material per Google’s guidelines” message or something. So the user knows instead of thinking that the bot is broken or taking a long time to post its message.
I did experience some mixed results (sub-par) with Gemini for summaries, is this only when NSFW content is in there? Could you share a screenshot or content?
also nice function, check for similar tags (either as a warning when you create new tags or as a weekly report to moderators to review which ones could be merged)
Just today I merged „German“, „Germany“ and „de“ or „acme-component“ and „component“ as a moderator. I can do that since I am just have a feel for most of the tags as I wrote a list but once the site gets more active and more moderators do part time jobs…
