I am working on a new (default disabled) persona for Discourse AI that specializes in deep research.
The key differentiator is that this persona is given permission to work through very large bodies of text, this unlocks a large amount of previously extremely complicated use cases.
Today I was looking at our main Kanban topic, it has gotten quite enormous and splitting it off into sub topics (features / bugs / support) can take an enormous amount of manual work.
Here is a link of how I leaned on Forum Researcher to help prep for the manual work:
It is a prime showcase of how this kind of tooling can assist us in working through some of the more complex parts of managing a forum.
Happy to take questions, the feature itself is now in Discourse AI but the persona is default disabled and I am holding off on documenting for a bit more while I refine the feature.
Hope you all are as impressed as I am at the general usefulness of this tool.
Breaking down those long-tailed topics was pretty brutal. Something like this would have helped immensely.
I found some advice I wrote about the manual way (though I appreciate that there are multiple ways to tackle the job) and thought it could be fun/useful to compare against the AI assisted version:
A feature request I have been considering is making this an allowed by group setting, so we can let our TL3 members reset bump date. It happens alot on meta, mostly because of spam but also because of topic gardening like in this case.
Iāve assigned this to myself and will be looking at it more closely this week.
Thanks for looking into this, Sam! Yes, I am impressed.
An individual topic shouldnāt be too bad. But having some extra tools on hand to tackle all the other outstanding ones too may tip the balance and help speed the task up (probably worth another āā for extra luck )
Though I think my main advice for any newer sites reading this is, āavoid creating kitchen sink topics in the first placeā. Improvements to the cure is nice, but I think prevention is still the best option if you can nip it in the bud early enough.
I deleted or moved all the replies from the kanban topic, so it has no replies currently.
The next step for the topic will be to update the first post which is also a big task! Itās gotten quite long and has outdated screenshots and instructions, instructions are technical and a bit confusing, and does not explain how to set it up for various purposes. If anyone wants to help out with this, let me know!
The forum helper was helpful. In the end however it is still a manual process of reading each post and deciding:
keep as a new related topic (still relevant)
delete and include in first post (still relevant, likely common ho)
just delete (not worth keeping/no longer relevant)
@sam I noticed alot of duplicate posts - would be interesting to see if AI can help identify these. These should be easily identified because they are from the same user and typically repeat the same questions in the same topic or a new topic.
Iām now looking at the ActivityPub Plugin topic, and asked the forum helper about it. The topic is a bit out of control but is also not an official plugin. @angus do you have thoughts on how youād like to tidy it up and does this AI report help?
Honestly, my current inclination is to keep the replies for archival purposes, but minimize them somehow so the topic feels a bit less intimidating. Most of the replies are just no longer relevant for the current plugin, and if folks read them for any practical purpose, i.e. using the plugin, then they may get the wrong impression.
I think deleting the replies would be regrettable as itās nice to have a record of the pluginās development over time, but if there were some way to easily minimize them by default, that would be ideal.
We are already handling ActivityPub support and documentation in other topics, and the OP already reflects that in its brevity and use of links. The plugin topic itself is more of a ācatch-allā for folks wanting to discuss/ask about the plugin.