Even better, make the responses opt-in or remove them entirely!
Or at least warn users that they are about to interact with a chat-bot when posting here.
It might seem easy to argue that it reduces time spent on trivial issues, but this is technical software and I’m sure most people using it are aware of LLMs, and if they wanted to ask one for an answer, they would. It’s free.
By deciding to ask here it’s likely they have made the decision to get human assistance, be that for ethical reasons, or maybe just because they have already asked their sentence-generator of choice and exhausted its (extremely limited) usefulness.
I’d find a chat-bot being the first response to one of my posts to give me significant reason to reconsider how seriously this project takes its users, and from reading a few posts here, it seems in practice to just end up with posters getting frustrated at best.
5 Likes
I completely agree with your premise that even better the make responses optional inside or remove them goblins.
1 Like
Hi, I agree, most of the value of a forum is to discuss with other users, but if a bot (clearly indicated as bot) post a reply, at best it’s useful and worse it’s a bit of noise and fellow users can still jump in and hopefully come with a better answer
and for what I see, the LLM has been RAGed with the latest doc version, I found it less bad than regular LLMs, but I didn’t look at many replies yet
ie: not a big deal, but a bit of clearer communication (like adding to the post template “an LLM bot will probably try to reply first, but the forum is full of real humans that you will be able to interact with”)
1 Like
The opt-out concern is fair, and we’ve made a change.
Some context on why AI replies exist on the forum at all: we’re a small team, and during busy stretches it can take us up to a month to get to a community question. Rather than have posts sit unanswered for weeks, our AI assistant has direct access to the latest Stalwart documentation and source code (it can search the actual stalwart, webui, cli, and website repos), so it can usually point people to the right doc page, config option, or code path while they wait for a human. Every AI reply is reviewed by a human afterwards, who will either confirm it, correct it, or follow up with a fuller answer.
Nevertheless, the point about wanting a human’s eye specifically is a reasonable one. As of today, we’ve added:
- a
no-ai tag: applying it to a topic suppresses the AI reply entirely
- a note in the composer on Questions and Issues explaining how to opt out
So if you or anyone else prefers to wait for a human, the option is there. Thanks for the nudge to make it explicit.
1 Like
This is a really awesome change, thank you so much for listening <3