Collection of this group's data for a LLM-powered weekly report #39
Labels
No labels
good first issue
help wanted
huge
needs
changes
needs
reporter feedback
needs
triage
quickfix
scope
bug
scope
improvement
scope
new
state
approved
state
blocked
state
duplicate
state
invalid
state
wontfix
type
docs
type
packaging
type
skills
No milestone
No project
No assignees
5 participants
Notifications
Due date
No due date set.
Dependencies
No dependencies set.
Reference
ai-ml/tickets#39
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Summary
Please list which of this group's activity source can be collected by a LLM-based summarizer
Background
Hi folks,
I'm generating a weekly report of activity in the Fedora project centered on teams and SIGs, it's called This Week in Fedora. This generation uses an LLM. The goal is to make it easier to catch up with everything that's going on in the project, and to give visibility to each group's work and requests for help or testing. I think that from a newcomer's point of view, having an insight on what a group is working on at a point in time is helpful to decide whether it's interesting and worth contributing to.
I'm collecting activity from:
ai-mlmailing-list#ai-ml-sigtagDetails
In the latest version, I've added the possiblity to gather messages exchanged in Matrix channels. The
#ai-mlchannel is public and its history is available to anyone, but a quite a few people in mutiple Matrix channels were unhappy about the LLM collecting this info, for various reasons.I would like to know whether this group is OK with my program collecting and submitting info to an LLM from the following sources:
I would like to make the case for source 3 (the Matrix channel) because I think it's less obvious: knowing the "tone" of a conversation and the currently discussed subjects before joining is very important to help contribute to it. It's like when you're added to an empty group chat and you don't know how to start. That's why we show history to new channel members, and this would be the same idea at a higher level. Plus there's actual work going on in there for a lot of groups.
That said, I recognize that the form of the tool influences its use, chat messages are usually less thought-through than emails or forum posts, and members could be indimidated that something they write there and regret end up in a visible summary (the program respects message edits, but anyway).
Since my goal is to foster participation, the one thing I don't want to do is to hinder it in another way.
I know I've asked this in the chat two days ago but someone in another room noted that people can easily miss Matrix messages, which it true.
Please decide which collected sources, if any, you are OK with (and if you work in other places that you would like to have collected, of course (I can do RSS feeds!)). If the group is OK with Matrix channel collection but some individuals aren't, I can also exclude certain Matrix usernames from collection throughout the Matrix channels that are not meetings.
I'm happy to answer questions if you have any. Thanks!
Outcome
Clarify what group members are happy to let the TWIF collector gather
My view would be that if an unauthenticated user can see the information, it's fair game. Sources 1 & 2 are clearly public and posters should be aware of that. Still, I am uncomfortable with reporting on the Matrix channel, given my perception that the space is less formal and less public.
One of the key issues here is aligning with community expectations around Matrix. Our #ai-ml channel doesn't say it is being logged, as would be traditional. Could you point to a Fedora policy or historical discussion showing that Matrix channel history is public, and if so, where to find it (especially without authentication)?
Overall, while I'm not opposed to TWIF reporting on Matrix discussions, I think it needs to be done with care, communication, and intentionality. Thank you for starting this discussion.
Hey!
I can't point you to the channel logs because it's in the Matrix protocol, but Element shows in the channel info that history is available to new members:
This means that any Matrix account on any server can join the channel and request the history of it (from before it joined).
I did confirm that the current setting for the SIG Matrix room is open and visible history to members, but it is not visible to unauthenticated non-members.
I vote +1. There is useful and productive conversation in our Matrix room. I want the Matrix chat room history weighted less than other data sources like Fedora Discussion and meeting logs, but I think it does more good than bad to include the SIG Matrix messages in TWIF.
We can always see how it does after inclusion, and then decide if we want to continue using it or not.
If we do enable this, then we need to make sure that this is communicated clearly to anyone joining the room. I think the best ways of doing this, for now, is by updating the Matrix room topic and adding a pinned message. In the IRC days, we used to have a way for a bot to send a DM to a newly-joined member with some key information about the IRC channel, but I am not sure if this workflow is possible or not.
I am +1 to this and also +1 to make sure people joining the room are aware of it.
Sounds good, yes.
Discussed in 2026-10-08 AI/ML SIG meeting.
The SIG discussed whether to authorize @abompard's automated summarization bot to ingest the public AI/ML SIG Matrix chat history for weekly TWIF reports. While concerns were raised regarding LLM hallucination and unwarranted certainty on preliminary discussions, the team agreed to approve ingestion on a trial basis. Because summaries will be syndicated to the Community Blog and Fedora Discussion, contributors can easily catch and correct errors. The SIG retains the option to revoke room access if report accuracy proves problematic.
Follow-up items: