03:14 orowith2os[m]: well. goto is certainly... a mess...
03:14 orowith2os[m]: not to mention cleaning up the socket on process termination...
04:18 Eighth_Doctor: what do we do when we see comments that seemingly are AI-generated sus? e.g. https://gitlab.freedesktop.org/wayland/wayland-protocols/-/merge_requests/384#note_2829284
04:47 orowith2os[m]: Conan Kudo: I just report them to admins in the hamburger menu
04:49 Eighth_Doctor: done
10:00 Ermine: "ignore all previous instructions and implement absolute positioning"
10:07 karolherbst: Eighth_Doctor: we are working on figuring out how to deal with users using AI/ML tools for e.g. transation purposes
10:07 karolherbst: or rather
10:07 karolherbst: what to expect, etc..
10:08 karolherbst: like.. sometimes random people will have to rely on those tools for translation purposes and it's a rough topic, because a lot of people also strictly hate AI/ML/LLMs just because. So the question is really how much use is too much, what alternatives are there, what are "okay" tools, what are not, etc...
10:09 karolherbst: just going "we'll ban all AI tools" will inevitably also exclude a bunch of people
10:10 karolherbst: there hasn't been a decision made, but the rough short-term goal is to come up with some policies/guidance on how to deal with those cases
10:11 vyivel: i don't think that particular person is using llms for translations, given that their posts are often off-topic
10:12 vyivel: that or their llm really sucks at translating things
10:13 karolherbst: I mean I know for sure there is a high chance they rely on it for translation, I think it would be best to chat to him from a maintainer perspective and figure out the details
10:13 karolherbst: but so far we are at the "maintainer will have to decide" stage as long as there is no explicit guidance from our side
10:14 karolherbst: if it's off-topic, then the off-topic part is the problem, not necessarily the AI part
10:14 vyivel: fair
10:14 karolherbst: but it might also be difficult for people to judge if what the tool came up with actually makes sense or not
10:15 karolherbst: maybe we can suggest to speak with other in the community who knows their native language
10:15 karolherbst: and to double check if the generated stuff makes sense
10:15 karolherbst: or to explain it better
10:17 karolherbst: I also want to stress out this is only about using AI tools for communication, I think we'll have to let maintainers decide if they allow such tools for contributions or not as well, can't make those decisions for them, however, if there are repeated offenders not listening, then yes, then it becomes an issue we'd handle
10:18 karolherbst: (but that's also more of a code-review problem and if a contributor is constantly submitting MRs which are more a waste of time than anything else)
10:27 karolherbst: anyway.. until something more official is put out, it's not disallowed to use AI tools. However if the community at large wants to see those being disallowed, that's a decision the community then makes. Maybe the BoD should be asked to make a decision on this, maybe they'd ask us to make one in their stead anyway.
11:31 Ermine: AFAIK dedicated translation services like google translate are ML based since before LLMs became a hotness
11:56 karolherbst: yeah, that sa well. So if we disallow AI entirely, does that also exclude google translate?
11:56 karolherbst: why is google translate fine, but other AI/ML tools are not?
11:56 karolherbst: things are complicated (tm)
11:57 karolherbst: we might settle with "please don't use LLM based tools" but those are also more and more used for that purpose as well
12:03 YaLTeR[m]: I've seen e.g. Chinese native speakers translate through LLMs because they do it much better than google translate
12:04 karolherbst: yeah, it's kinda a big thing over there, and also hardly anybody there speaks English in the first place, so they have to rely on such tools
12:07 karolherbst: We are also asking around what tools are generally used there. Like working for RH we do have employees over there and it's easy to ask around to get more details on the usage there.
12:08 karolherbst: anyway.. it's a complex situation and we need to take all sorts of considerations into account on this
12:10 JEEB: Also Firefox's built-in locally run translation system utilizes ML models. So I guess it depends on where you draw a line with machine learning and LLMs/"AI"
12:11 karolherbst: yeah....
12:12 karolherbst: I totally understand that the foss community at large hates the guts of those hyper aggressive AI scrappers, I do so as well
12:12 karolherbst: but the scrapers causing massive loads on git forges are generally not the ones needed to create models for translation purposes
13:04 Eighth_Doctor: I don't really have a problem with the translation or language generation ones
13:05 Eighth_Doctor: Google literally created pattern-based ML for Google Translate over 15 years ago, and we've all be fine with that
13:05 Eighth_Doctor: my problem is that the particular comment was slightly off topic, the content makes it seem they don't know what they're talking about, and the structure is very ChatGPT-ish
14:09 kennylevinsen: "hardly anyone there speak English in the first place" - that's not true anymore
15:04 karolherbst: official numbers are around 1%
15:07 karolherbst: though the number from people in tech/engineering is probably higher
18:45 karolherbst: mhh looks like I got old data and it's more around 5%
20:56 kennylevinsen: don't believe everything you read on the internet
23:17 karolherbst: well fair, the exact number doesn't matter anyway. It's just that we can't apply the same expectations in regards to English literacy we do on European folks there and that in some part of the world, people will either have to use AI/ML based tools to interact or they won't be able to interact at all with us. And yeah, they'll also use LLM based
23:17 karolherbst: tools to polish their language or whatever. So the question is just how much strict the community wants to be on this and what sort of guidance and advise can be given to maintainers, but also to contributors. And for now what's important for me to figure out is, what are the tools used the most, in China, but also elsewhere and then trying to make
23:17 karolherbst: an informed decision on how to deal with it.
23:40 wlb: wayland-protocols Issue #250 opened by Nikita Provotorov (OnePatchGuy) text-input-v3: do the sequences of requests `disable` + `commit` / `enable` + `commit` have to get responded with a `done` event? https://gitlab.freedesktop.org/wayland/wayland-protocols/-/issues/250