[10:47:05] 10ORES, 10Scoring-platform-team (Current), 10Puppet: orespoolcounter1002.eqiad.wmnet reporting compile errors - https://phabricator.wikimedia.org/T213586 (10akosiaris) 05Open→03Invalid All are warnings, that is not errors and are safe to ignore. They are about a feature (exported resources[1]) that is no... [10:47:06] El búfer 1 está vacío. [15:04:25] o/ [16:33:48] 10Scoring-platform-team, 10Analytics: Investigate formal test framework for Oozie jobs - https://phabricator.wikimedia.org/T213496 (10fdans) p:05Triage→03Low [17:03:57] harej: o/ [17:04:04] harej: We are lacking our noble chair [17:35:47] 10ORES, 10Scoring-platform-team (Current): ORES command line service sometimes hangs - https://phabricator.wikimedia.org/T205909 (10Halfak) a:03Halfak [17:36:08] 10ORES, 10Scoring-platform-team (Current): ORES command line service sometimes hangs - https://phabricator.wikimedia.org/T205909 (10Halfak) https://github.com/wikimedia/ores/pull/309 [17:44:09] 10ORES, 10Scoring-platform-team, 10Documentation: Document/script ORES config change deployment process - https://phabricator.wikimedia.org/T213743 (10Halfak) [17:44:47] 10ORES, 10Scoring-platform-team (Current), 10Patch-For-Review, 10Puppet, and 2 others: ORES services should bind to ores config files - https://phabricator.wikimedia.org/T210719 (10Halfak) Maybe we should have a script and a process instead for manually restarting ORES nodes in a safe way. See {T213743} [17:59:01] Well, tire place has an availability so I'd better do this now. Hope to be back in ~2 hours [18:20:02] Good luck. I hope it works out. [18:54:45] Just finished partnership for AI call. [18:54:47] Heading to lunch [18:58:29] Let me know how it went when you’re done lunching. [19:57:38] harej, yeah, not so much faith in this working group generally. I think they have gotten a good group of people together. [19:58:17] But there's a constant push for actionable output yet most people on the call are discussing something different. Getting on the same page ourselves is essential before we can produce any recommendations for the rest of the industry. [19:58:52] Most of the work right now is in producing case studies that exemplify important aspects of AI/human collaboration. But even the definition of that is weird and different between people. [19:59:16] It seems I'm the only one who is really focused on collaboration beyond mere use of an AI. [19:59:39] I want to talk about how people's values get enacted into an AI, how it is developed, used, regulated, etc. [19:59:55] So I want to talk about societies, cultures, and social systems. [20:00:49] Those points of view are not really shared. :( So I can invest a bunch of time and try to champion them or keep my interactions light and let the single-user collaboration scenario be the dominant exploration. [20:21:00] It's hard for me to see someone using a Product in the tool-user scenario as "collaboration". I don't collaborate with my hammer. But I might collaborate with a tool builder to build a better hammer or design a new tool. [20:21:56] Even when "AI" is involved, I still think that such tools are closer to a hammer than another human. [20:22:23] So long as I'm using the AI as a //means//, we're not really collaborating. [20:23:27] The vision of product teams developing AIs for users seems to have become the new "view from nowhere" for AI discussions. [20:23:53] It's taken for granted that there will be specialized product teams building AIs and users merely using. [20:24:01] Maybe giving feedback [20:24:47] It's times like these when I realize that our world of open collaboration is very unusual. [20:50:40] Nice that we found that workshop at CSCW problematizing exactly this question of participation. [20:52:54] halfak: I'm kinda parachuting in here but I think those questions ('how do values get enacted, how is it used/regulated') are the interesting ones [20:53:07] but people need to have a little bit of grounding before they see that context [20:53:14] awight, agreed. [20:54:07] I wonder if there's a way to frame the discussion (maybe making participants do some reading first0 to get them 'up to speed' as it were [20:55:22] apergos, yeah that seems like a better model of approaching this. Essentially, I'd need to shift to workshop organizer mode. [20:55:37] And build a format that would help people explore those kinds of questions. [20:55:42] apergos: fixed structures determine most of what can be done or expressed, and then you have to consider the structure deployed to create the structure... [20:56:22] I think structure can be generative when not oppressive. But mindsets are harder to deal with. [20:56:53] You get questions like, "How can we help the user understand this?" It takes a lot of energy to push back on the idea of "the user" as being problematic. [20:57:00] It's almost as if a "partnership" started by Amazon, Facebook, Deepmind etc would be resistant to change [20:57:26] ^ I think that is a real thing. [20:57:49] I think there is a lot of goodfaith involved. But also a single culture is represented when it comes to imagining how AIs might work. [20:59:04] imagine that, resistant to change [20:59:10] but also honestly [20:59:24] people do have a sort of monoculture that is only slowly evolving [20:59:30] around this stuff [21:01:27] sorry, I'm gonna be AFK for a bit--my hot water heater is flooding the kitchen [21:04:16] Oh no! That sounds bad. Good luck. [21:04:28] Speaking of luck, how did the tire stuff go awight? [21:07:19] I've been thinking about diffs in Jade. If we have a fancy view, we're going to need a fancy diff. [21:09:04] The good news is that I think a bit of logic can go a long way and be very performant. We have a relative subset of actions that we need to represent. [21:09:46] The biggest issue, I think, is going to be around representing an arbitrary action performed against the JSON. [21:09:57] We might need to render quite a lot in two columns. [21:11:46] 10Jade, 10Scoring-platform-team (Current), 10Design: Jade Wireframes: Entity edit mode - https://phabricator.wikimedia.org/T212374 (10Halfak) a:03Halfak [21:11:50] good luck awight, hope the flood is contained [21:11:56] 10Jade, 10Scoring-platform-team (Current), 10Design: Jade Wireframes: Entity view mode - https://phabricator.wikimedia.org/T212379 (10Halfak) a:03Halfak [21:17:14] :) perfectly happy to be a renter, today [21:18:29] halfak: Where are you rendering JSON diffs? I didn't think that would appear in any of our mockups. [21:18:45] The diff should be of content, not of the judgment edit... [21:18:58] https://phabricator.wikimedia.org/T212377 [21:19:19] I haven't started yet. Just thinking about it while I work on the edit view. [21:20:42] 10Jade, 10Scoring-platform-team (Current), 10Design: Jade Wireframes: Entity edit mode - https://phabricator.wikimedia.org/T212374 (10Halfak) https://docs.google.com/drawings/d/1fP-4Tmn4bDDLWKD_Uu_W0xQNM-QyRc4LDKQTkKNp_fc/edit covers a few different types of edits someone might do. Menu items allow "edit",... [21:21:35] awight, ^ some attempts at editing. [21:21:38] harej, ^ [21:21:41] halfak: ah yeah so that's a content diff, not a judgment diff [21:21:59] Oh I see. We'll need judgment diffs though [21:22:37] They already work pretty well, (/me scrounges to demonstrate) [21:24:22] https://en.wikipedia.beta.wmflabs.org/w/index.php?title=Judgment:Diff/376901&diff=387140&oldid=387139 [21:24:33] JSON is line-based, which is a good match for the default renderer. [21:25:09] Here's the fancy sauce if we want to keep up with the Joneses, https://www.wikidata.org/w/index.php?title=Q60379359&curid=60252871&diff=834482508&oldid=834476423 [21:25:25] but IMO this is for a later phase [21:25:34] Right. I think this is definitely better than the JSON-based line diff. [21:25:45] But I agree that the JSON-based line diff is actually quite good. [21:32:29] hehe "full name Partnership on Artificial Intelligence to Benefit People and Society" [21:32:52] that's when you know there are definitely skeletons in the closet. [21:41:08] halfak: If you agree with the concept? https://github.com/wikimedia/ores/pull/310 [21:42:10] awight, I'm conflicted. I feel like there are better options for turning a stream of revids into JSON blobs. [21:42:24] E.g. tsv2json or sed. [21:43:08] Maybe they are running in an environment where those aren't good options? [21:43:11] mandatory {"rev_id": 123456} JSON is an odd choice though? [21:43:52] awight, the output will be JSON regardless, right? [21:43:54] I see the rationale, that you might want to merge input records with score data... but that seems like the special case rather than the most common need. [21:44:00] For now yes [21:44:14] +1 I'll just recommend tsv2json to them. [21:44:36] I get tsv2json from the python package "json2tsv" [21:45:10] cat my_revids.tsv | tsv2json int | ores score_revisions > my_output.json [21:45:12] very nice [21:45:32] I think u need the field name, but I get the point. [21:46:19] Only if there is a header :) [21:46:26] Actually, no. It's ordered. [21:46:56] (I wrote the script and made massive improvements to this library) [21:46:57] :) [21:47:20] okay header is required, I see. [21:47:44] that's not optimal but they can deal [21:48:15] Right. Not many options in the script. It's relatively easy to add a header to a stream. [21:48:33] echo "rev_id"; cat my_revisions.tsv | ... [21:49:14] ^ Missed some parens [21:49:24] (echo "rev_id"; cat my_revisions.tsv) | ... [21:50:06] $ man dog --> "No manual entry for dog" [21:50:10] How did this happen? [21:50:17] I though dog was man's best friend! [21:50:21] * halfak shows himself out. [21:53:39] 10Jade, 10Scoring-platform-team (Current), 10Design: Jade Wireframes: Entity edit mode - https://phabricator.wikimedia.org/T212374 (10Halfak) My thought is that "move" only shows up for one's own endorsements. It's really just a convenience since "remove" and "endorse" perform roughly the same action. Edi... [21:56:39] I'm second-guessing bringing Jade to an Activities meeting this month. [21:57:00] On one hand, we can bring more attention during the development process which I think is a win. [21:57:11] On the other hand, Jade is hard to explain. [21:57:41] Maybe I could position this as a discussion of how AI is just different around Wikimedia than it is anywhere else -- where Jade is part of the story. [21:57:50] harej: ^ [21:58:30] halfak: maybe this initial presentation can be around more of the fundamentals: things that haven’t really changed since we started working on this [21:58:57] Not sure I see what you're getting at. [22:02:19] In any case, I’m sympathetic to the idea of holding off on presenting [22:08:38] Gotcha. I'll think about it more. I do have a talk I'm half-working on around how we do AI differently -- in many ways without trying. [22:08:57] My original thinking was "if there are things we haven't done yet we can just show the things we did do!" and then I realized that amounts to slim pickings for presentation purposes [22:10:36] This isn't a live demo though, we can show people a wireframe, or the current rendering on beta [22:12:24] awight, that's a good point. [22:12:29] It might be helpful to recap what our goals are in giving an all-staff presentation? [22:12:31] By next week, I think we'll have a lot of wireframes. [22:13:16] awight, that's a good point. There's lots to teach this audience about WRT coordination between patrollers/workflows and how our users audit ORES today. [22:13:55] Oh here's another glitch: Is there really a metrics meeting? the 31st would be during All Hands, and I don't see anything on the calendar. [22:15:06] hmm it's on the meta page, https://meta.wikimedia.org/wiki/Wikimedia_monthly_activities_meetings [22:15:46] strange, not on the WMF staff calendar or "fun and learning" [22:16:13] it might be worth contacting ... somone. [22:17:04] Yeah. I assumed it was on the 31st in order to take advantage of allhands. [22:20:13] halfak: in -office, varnent is suggesting it will be moved [23:25:08] I'm off to the gym. Have a good night folks. [23:25:24] 💪 [23:37:16] 10Scoring-platform-team: Formalize fallback rules for automatically determining ORES threshold levels - https://phabricator.wikimedia.org/T213772 (10awight) [23:37:31] 10MediaWiki-extensions-ORES, 10Scoring-platform-team: Formalize fallback rules for automatically determining ORES threshold levels - https://phabricator.wikimedia.org/T213772 (10awight) [23:38:16] RoanKattouw: ^ fyi [23:44:03] 10MediaWiki-extensions-ORES, 10Scoring-platform-team: Formalize fallback rules for automatically determining ORES threshold levels - https://phabricator.wikimedia.org/T213772 (10Catrope) The defaults are as follows, but I eyeball the results and adjust where needed. For example, if the 90% precision level has... [23:45:53] 10Scoring-platform-team: Add table view to ores-reference-ui - https://phabricator.wikimedia.org/T213773 (10awight)