XSF Discussion - 2023-10-30


  1. MSavoritias fae.ve

    > but now that i think about it, it would be probably a very big XEP Depends on the scope and the function. And how we want to approach it. Extending ad hoc commands to recognize a group of muc's and manage moderation for them is probably doable.

  2. MSavoritias fae.ve

    And chogram already has "related" muc's which are close to that

  3. Kev

    Yes, there was a discussion at the last summit, but no-one else wanted to write it up and I've not found time (yet).

  4. pep.

    https://xmpp.org/extensions/inbox/xep-ai.html Yet another spec that makes it sound like I need to opt-in to opt-out? > The messaging app should provide a clear indication that the suggested messages are generated by an AI service, and provide an option for the user to opt-out of the feature.

  5. pep.

    Can we get rid of these please? (also can we get rid of AI BS too plz)

  6. Zash

    Does this fill a hole in XMPP specifcations?

  7. Kev

    Council has the opportunity to veto such things, my only job is getting it from my mailbox into the inbox (and I failed to even do that in a timely manner)

  8. pep.

    Zash, is that a hole the XSF is interested to fill, that's the real question

  9. pep.

    (I wouldn't)

  10. emus

    > pep.: > 2023-10-30 05:57 (GMT+01:00) > https://xmpp.org/extensions/inbox/xep-ai.html Yet another spec that makes it sound like I need to opt-in to opt-out? Thanks, side-topic in general: Anything around this would be an apprecitaed topic at GSoC

  11. emus

    (if its not vetod)

  12. singpolyma

    Even if it is. Nothing says we need a xep number to get some work done 😉

  13. Zash

    This fills a hole in the marketing strategy?

  14. pep.

    This certainly fills a hole in the way to attracting VCs

  15. emus

    > Zash: > 2023-10-30 06:10 (GMT+01:00) > This fills a hole in the marketing strategy? Maybe, but I am not a fan of AI I think. But I also don't stand in the way if anyone believes they want do something around it

  16. pep.

    go go gadget AI

  17. nicoco

    pep. I wrote it in the ML (if it works?), my main concern with this XEP is that it calls AI what really is "a conversational agent"

  18. singpolyma

    Nah, gadget is all artificial *except* his intelligence

  19. singpolyma

    nicoco: well, AI is way too broad a term to be meaningful. Any heuristic algorithm of any kind counts as AI

  20. pep.

    > Nah, gadget is all artificial *except* his intelligence true :)

  21. pep.

    Apart from that, I did have a concern that can be addressed by the council at some point (dunno if that's protoxep or experimental) > Yet another spec that makes it sound like I need to opt-in to opt-out?

  22. singpolyma

    I didnt read it yet, but generally opt in/opt out semantics are UX and should be out of scope for a XEp

  23. singpolyma

    I didnt read it yet, but generally opt in/opt out semantics are UX and should be out of scope for a XEP

  24. pep.

    393 is another contender that has the same issue fwiw

  25. Zash

    Stochastic programming®

  26. nicoco

    pep. I agree that it feels a little "let's write AI here so we show we are not outdated", but I don't think it's necessarily bad to have a spec allowing better interfaces to interact with bots in general. There is a middle ground between "falling for the hype" and not recognizing that there are use cases for machine learning.

  27. pep.

    singpolyma, also I'm more and more convinced you can't do specs without having UX in mind already. Maybe that needs to be acknowledged somewhere someday

  28. singpolyma

    Well, you need to have at least one UX implemented already IMHO, to write a spec, but I know not everyone agrees

  29. nicoco

    I don't understand your concerns about opting out tbh pep. Right now, you might already be talking to large language models without knowing it (I may be skynet in disguise). This spec gives you a way to opt out?

  30. nicoco

    > nicoco: well, AI is way too broad a term to be meaningful. Any heuristic algorithm of any kind counts as AI I couldn't agree more, I hate the term AI for various reasons but hey, now it more or less means "artificial neural networks" or "deep learning". It's the way we use words that define what they mean. I also hate when old grumps say "this word shouldn't be used to mean this because in my time it meant that", so I try to act consistently.

  31. pep.

    nicoco, it's like the #norobot on Mastodon I guess. I'd like it the other way around. "Please don't send me crap as long as I haven't expressed interest for it". Surely I could already be deceived and then, well too bad, but if I have a choice I prefer to be asked first

  32. pep.

    You know, consent and all :)

  33. nicoco

    I'm not very familiar with it but I think #norobot means "don't scrape content I have produced to train your ML models", and I don't think this is similar to what we have here.

  34. pep.

    I'm not saying it's similar

  35. pep.

    La forme pas le fond (unsure how to say that in english)

  36. pep.

    form/substance

  37. pep.

    I'm saying I'm tired of having to implement thing just to opt-out of them

  38. pep.

    As a developer. And as a user, I want to be asked consent first

  39. nicoco

    Anyway, I agree that this optin/optout part is weird, since the XEP describes a way to "query a model". I guess the intended use case is "automatic reply suggestion" as we have in a some mail/IM clients these days.

  40. pep.

    Maybe that's just the way it's being phrased. Probably you'd send a message first to get a reply.

  41. pep.

    But this "opt-in to opt-out" is a pattern we see here and there

  42. emus

    (I cannot tell if there is something meaningful, but I would not just see the AI & XMPP topic only within a chat context but also IoT & RTC)

  43. emus

    Dear community, please remind to sign-up for the XMPP Summit 26 or at least tell that you are considering to join (and maybe reasons that prevent you). That very much helps organisation. https://wiki.xmpp.org/web/Conferences/Summit_26#Summit_Participants If you have no access please ask the members to help you. Cheers, Eddie

  44. moparisthebest

    emus: do we have an idea of what time+timezone the remote participation would be?

  45. Daniel

    The summit is an all day event in European time zone. (utc+1) Probably from 10am to 5pm or so

  46. Zash

    Isn't it +2 by then?

  47. Zash

    or

  48. Zash

    no

  49. Zash

    This is my brain on DST switchover

  50. Daniel

    There is no special slot allocated to remote participation but instead we try to 'stream' the whole event and remote participants can signal that they want to speak

  51. Daniel

    That said if you have the time and money to participate in person I reckon that's much better. A lot of summit is just getting to know people and discussions over lunch and dinner

  52. moparisthebest

    4am - 11am for me, oof

  53. Zash

    And reminding ourselves that there are actual humans attached to these nicknames :)

  54. Daniel

    The summit usually consists of multiple discussion topics if you can indicate your interest in certain topics beforehand we could maybe try to schedule them towards the end of the days

  55. Daniel

    Topics are proposed, sorted and scheduled on the morning of day 1

  56. MSavoritias fae.ve

    > https://xmpp.org/extensions/inbox/xep-ai.html Yet another spec that makes it sound like I need to opt-in to opt-out? >> The messaging app should provide a clear indication that the suggested messages are generated by an AI service, and provide an option for the user to opt-out of the feature. This looks like a horrible can of worms to open for xsf the council and the xmpp ecosystem. The fact that is has this mentioned: > Asking artificial intelligence to make predictions based on a past conversation Without consent and privacy considerations gives me pause.

  57. MSavoritias fae.ve

    > I don't understand your concerns about opting out tbh pep. Right now, you might already be talking to large language models without knowing it (I may be skynet in disguise). This spec gives you a way to opt out? From what i can see you can only opt out from seeing generated messages nothing more: > The messaging app should provide a clear indication that the suggested messages are generated by an AI service, and provide an option for the user to opt-out of the feature. And i dont see why we should enable such behavior anyway imo

  58. MattJ

    It's not "enabling" AI, it's ensuring that when it *is* used there is a standard way to mark it up

  59. MattJ

    The alternative options are non-standard markup or no markup (the latter being the most likely outcome in most cases, whether this protoxep gets accepted or not)

  60. moparisthebest

    There are plenty of AI using XMPP bots now, you don't know they are AI until you talk to them and they spout nonsense back at you though :)

  61. MSavoritias fae.ve

    another alternative is also not enabling "AI". :)

  62. MSavoritias fae.ve

    exactly. whether this gets accepted or not changes nothing in AI

  63. MSavoritias fae.ve

    what it does change is xsf position on it.

  64. Zash

    Are they going to be doing this anyway? Might as well have a XEP for it. Assuming it's really needed?

  65. lovetox

    ? someone wants to mark a AI generated message as AI generated

  66. moparisthebest

    It lets bot authors tag their messages as AI generated so users can ignore all AI messages if they wish, right?

  67. lovetox

    if xsf accepts this, no position is declared

  68. moparisthebest

    Or display them differently etc etc

  69. lovetox

    xsf has no position on what users hook up to their network, AI or not

  70. MSavoritias fae.ve

    is it? because it seems to me that if this gets accepted xsf says AI is ok

  71. MSavoritias fae.ve

    which I am not saying is a wrong position. but it is a position

  72. MattJ

    XSF supports proprietary networks by publishing specs about gateways?

  73. moparisthebest

    AI is ok

  74. MSavoritias fae.ve

    pretty much yeah

  75. moparisthebest

    My ai bot uses ai defined in 1906

  76. MattJ

    I think it's best we leave out of the discussion whether or not AI is okay

  77. MSavoritias fae.ve

    xsf says its ok for people to communicate with discord

  78. moparisthebest

    Complicated if/else is AI

  79. MSavoritias fae.ve

    > I think it's best we leave out of the discussion whether or not AI is okay i didnt mention it

  80. MSavoritias fae.ve

    i said that its the position xsf would take

  81. lovetox

    XSF takes no position what is "ok"

  82. MattJ

    No, moparisthebest did, and I used my inbuilt LLM to predict the rest of the conversation and it wasn't good :)

  83. MSavoritias fae.ve

    if we are ok with it we can go ahead

  84. MSavoritias fae.ve

    ah

  85. moparisthebest

    A server deciding how to route message stanzas is AI

  86. lovetox

    people can do with their servers what they want

  87. MSavoritias fae.ve

    either way my problems with this xep are more than the AI

  88. MSavoritias fae.ve

    like the fact that it has no privacy or consent considerations at all

  89. moparisthebest

    That doesn't make sense in this context

  90. MattJ

    We don't have that for any other service you might interact with on XMPP either

  91. MattJ

    Like when you send a message to Discord

  92. MattJ

    In some cases there's not even much we can do about that at the protocol level - see matterbridge

  93. moparisthebest

    It's a bot sends a message to a muc, that already happens, this just let's the bot say "btw I'm a bot" that's all

  94. MSavoritias fae.ve

    but the difference is that the user explicity contacts discord

  95. MSavoritias fae.ve

    this xep mentions though

  96. MattJ

    and the same will be true of "AI" - bots *already* exist that use ChatGPT and such

  97. Zash

    XMPP approves of Embrace eXtend, and extinguish by supporting gateways!!!

  98. MSavoritias fae.ve

    > 4.3 Asking artificial intelligence to make predictions based on a past conversation

  99. MattJ

    In such a case a user submits stuff to a service, and the service responds

  100. MSavoritias fae.ve

    which afak gateways dont do no? in the sense that, they dont get random conversations from other people and submit them to discord

  101. Zash

    Quickly, someone feed all of logs.xmpp.org into some markov chain thing!

  102. MSavoritias fae.ve

    and again

  103. moparisthebest

    My bot that uses Markov chains from 1906: https://code.moparisthebest.com/moparisthebest/ash One using llama: https://github.com/divestedcg/llamabot

  104. MSavoritias fae.ve

    if we are okay with AI and ok with the conversations of everybody on the entire network being fed into this AI

  105. lovetox

    MSavoritias fae.ve, you seem to misunderstand what the XSF does, you think it governs how people have to use xmpp servers

  106. MSavoritias fae.ve

    then sure

  107. MSavoritias fae.ve

    > MSavoritias fae.ve, you seem to misunderstand what the XSF does, you think it governs how people have to use xmpp servers not really no

  108. moparisthebest

    This doesn't enable sucking in all XMPP conversations or anything

  109. MSavoritias fae.ve

    moparisthebest, did you read this part? 4.3 Asking artificial intelligence to make predictions based on a past conversation

  110. MSavoritias fae.ve

    of the xep

  111. lovetox

    it just publishes a standard, thats like saying the TCP standard which defines how data is exchanged with the protocol should have a sentence in it "But dont send TCP packets without user consent, display something before"

  112. moparisthebest

    Consent is already built into core XMPP

  113. Zash

    Enter https://datatracker.ietf.org/doc/html/rfc3514

  114. MSavoritias fae.ve

    > it just publishes a standard, thats like saying the TCP standard which defines how data is exchanged with the protocol should have a sentence in it "But dont send TCP packets without user consent, display something before" not really. AI is not just transport. i suggest you read some papers from the dair institute to know more about the difference

  115. moparisthebest

    If you send a message to a user or group of users, you consented for them to read it

  116. MSavoritias fae.ve

    > If you send a message to a user or group of users, you consented for them to read it of course

  117. MSavoritias fae.ve

    but have i consented for that message to be sent to third party AI services?

  118. moparisthebest

    You can consent to accept a message from a user or not

  119. lovetox

    thats between you and your server operator , not the XSF

  120. MSavoritias fae.ve

    remember that the consent for recordings exists for that reason

  121. moparisthebest

    Who said anything about 3rd party ai services?

  122. moparisthebest

    The two AI XMPP bots I just linked are local

  123. MSavoritias fae.ve

    > thats between you and your server operator , not the XSF you seem to be missing what this conversation is about. Its this: if the xsf is okay with AI and ok with the conversations of everybody on the entire network being fed into this AI (according to xep)

  124. moparisthebest

    The XEP says nothing about that

  125. MSavoritias fae.ve

    of course it does

  126. MSavoritias fae.ve

    i have linked it twice now

  127. MSavoritias fae.ve

    please read the xep we are talking about

  128. moparisthebest

    It's not even possible to get conversations of everyone on the whole network

  129. lovetox

    it is neither "ok" with it nor "not ok" with it, it simply should not care

  130. lovetox

    what users and server operators do with data exchanged is their business

  131. MSavoritias fae.ve

    lovetox, then you propose not to publish it?

  132. lovetox

    how do you get this from what i said

  133. MSavoritias fae.ve

    > It's not even possible to get conversations of everyone on the whole network not at the same time no. but it does make every user on the network second guess who they talk to because xsf approved their messages may be fed into an AI service

  134. lovetox

    if i dont care about something, why would i *not* publush something, that would indicate that i do care

  135. moparisthebest

    >> It's not even possible to get conversations of everyone on the whole network > not at the same time no. but it does make every user on the network second guess who they talk to because xsf approved their messages may be fed into an AI service This is utter nonsense, no other way to put it

  136. moparisthebest

    Anyone can, and maybe already is, running a bot now in all public channels sucking data for whatever purpose they want

  137. MSavoritias fae.ve

    > how do you get this from what i said simple. we have three options: - publishe it. (xsf thinks ai is a positive thing. and doesnt mind amplifing its usage. and the privacy issues the xep currently has) - dont publish it (xsf is neutral) - publish that ai is not allowed in its CoC (xsf is against AI)

  138. MSavoritias fae.ve

    the publish gives a platform to it essentially

  139. moparisthebest

    If we approve or don't approve a xep that let's someone mark a message from a bot ai generated doesn't affect that in the slightest

  140. MSavoritias fae.ve

    moparisthebest, of course. never said otherwise. but is it with xsf approval?

  141. lovetox

    MSavoritias fae.ve, you misunderstand what the XSF is here for, but it seems im not getting through with my explainations, so i leave it for now

  142. MSavoritias fae.ve

    because if this xep gets published it will be :)

  143. moparisthebest

    XSF doesn't approve actions

  144. emus

    yes about this ^

  145. emus

    Will add thislater if I dont forget

  146. Arne

    Hi, sorry I didn't follow much lately. It's not possible to join from a monocles server, neither monocles.de nor monocles.eu .

  147. Zash

    Why? Is there an error message or something?

  148. Arne

    I just says server not found

  149. Arne

    just checking gajim console again..

  150. Arne

    yep, just address not found.

  151. lovetox

    check with the server operator

  152. pep.

    « XSF takes no position what is "ok" » my eyes hurt, again.

  153. Zash

    xmpp.org → monocles.de connectivity works

  154. Zash

    the other direction .. less so

  155. Zash

    can't see more than that from here tho

  156. Arne

    thanks. Then I check it again on our site. Though it suddenly happend

  157. Zash

    Maybe the operators@ room is more on topic for this

  158. Arne

    indeed

  159. taba2

    best ios client?

  160. taba2

    movim, snikket?

  161. Arne

    I would still say siskinIM, but monal is also gettting better and better

  162. taba2

    monal has an annoying ui bug they said would take too much effort to fix...

  163. taba2

    Arne: why siskin and not snikket

  164. taba2

    i read siskin has unusable defaults

  165. Arne

    Sorry, I don't use iOS actually. Maybe snikket is fine too

  166. Arne

    But last time I could try a iOS client siskin was working pretty good

  167. taba2

    does it have stickers a a/v calls

  168. taba2

    does it have stickers and a/v calls

  169. taba2

    and notifications

  170. Arne

    a/v calls and notifications did work, as long as your server supports it.

  171. Arne

    when you're using a snikket server I guess the snikket clients are recommended

  172. taba2

    docker goes against my religion

  173. taba2

    my abusive religious cult

  174. Zash

    Arne, assuming you have an easy way to test something, find another server _not_ using port 5269 and try reaching that

  175. Zash

    port 9269 is used here. sometimes people decide to block outgoing connections to ports other than 5269, which could explain the monocles issue