Hmm yeah, good point. We lost this:
Let me have a look.
2 Likes
I think waiting for more feedback is a wise choice for this case.
Yes, I could have explained the purpose of those two examples better. Thanks for asking for clarification.
If I had written “experimenting with this updated search” instead of “a use case I expected to work,” I think my intent would have been clearer.
I started out checking a few use cases, but somewhere along the way that turned into broader experimentation with how the updated search and Ask AI behave.
I think what I need to do next is take some time to create a state diagram so I can better understand the different states, options, and choices involved—hopefully including the relevant admin settings.
I have a mental picture of how the pieces fit together, but I want to ground that in the actual behavior and implementation rather than assumptions. Ideally, I can locate the relevant source code and prompts and then use Codex or Claude to help create the diagram.
For now, please do not spend any more time on that reply. You can leave the ball in my court while I investigate it further.
1 Like
I got Jarvis to give it ago ymmv, AI content, grounded on our implementation Discourse Ask AI — an interactive field guide
1 Like
Finally figured out Jarvis
1 Like
Sometimes I wonder why you let met test these things.
A user posted a new topic and asked a few questions in the topic.
I looked at Related topics but did not see one that looked of value.
The next item on this list was to have Discourse search look and use the AI.
As this is a new topic, offering “Create topic” is not a wise reply.
Related to same new topic, different search.
Maybe a change of the reply wording.
For this topic
the AI reply is not one I would pass on after human review.
Should the AI reply include Reinforcement Learning from Human Feedback (RLHF)?
Should RLHF be limited to select users. I suspect some users might subvert the replies using RLHF.
In checking the Admin settings, thanks to Jarvis, did not see one where the AI prompt could be customized locally.
In link below scroll down to What an admin can change.
For this topic selected the OP question
Now exploring Slack → Codex → GitHub as a cheaper approach. can you guyz tell me if anyone has tried this?
and gave it to Discourse search to use with AI
Notice the first item in the list of topics is the currently open topic.
Your call, but would think that for the short list, as opposed to the “Show all matching topics”, the current topic would not be included.
Same topic, instead of selecting the question then pasting into search, used a right mouse click to open the context menu
clicking “Ask AI”
I was just curious to see what would happen, just experimenting.
Maybe this context menu needs an additional menu item; your call.