Meta & Children's Online SafetyChat: it's literally just mathematically optimizing for vpvs and sessions for teens as with everyone else
Raw text
Case 4:22-md-03047-YGR Document 2889-23 Filed 03/27/26 Page 1 of 11 AMENDED Exhibit 214 PLAINTIFFS' OMNIBUS OPPOSITION TO DEFENDANTS' MOTIONS FOR SUMMARY JUDGMENT Case No.: 4:22-md-03047-YGR MDL No. 3047 In Re: Social Media Adolescent Addiction/Personal Injury Products Liability Litigation Case 4:22-md-03047-YGR Document 2889-23 Filed 03/27/26 Page 2 of 11 Message From: [/O=THEFACEBOOK/OU=EXCHANGE ADMINISTRATIVE GROUP (FYDIBOHF23SPDLT)/CN=RECIPIENTS/CN= Sent: 10/25/2023 2:33:22 AM To: Subject: Message summary [{"otherUserFbId":100021478472088, "threadFbId":null}] Attachments: 395904239_1571448040326703_7256468352372961683_n.png; California-v-Meta-complaint.pdf; 395664739_712304456952550_8315227145069148601_n.png; 396040808_393873692974528_8150676540219452764_n.png; 395412074_285748414366445_6618436928385208323_n.png (10/24/2023 09:39:53 PDT): >again I have to ask... Are the compliance people reading the same litigation and legislation that I am? https://www.nytimes.com/2023/10/24/technology/states-lawsuit-children-instagram-facebook.html (10/24/2023 09:40:09 PDT): PRIV (10/24/2023 09:40:20 PDT): >does I am asking the same question. (10/24/2023 09:40:56 PDT): >realize how the core algo is set up? Like it's literally just mathematically optimizing for vpvs and sessions for teens as with everyone else, which is the thing we're being told we cannot do... (10/24/2023 09:41:09 PDT): >So my answer is: something is going missing in translation because this is our press response. >we share the attorneys general's commitment to providing teens with safe, positive experiences online, and have already introduced over 30 tools to support teens and their families. We're disappointed that instead of working productively with companies across the industry to create clear, age-appropriate standards for the many apps teens use, the attorneys general have chosen this path. (10/24/2023 09:41:20 PDT): >Like nobody is asking us to build 30 tools. (10/24/2023 09:41:26 PDT): >right? (10/24/2023 09:41:34 PDT): >I had the exact same reaction this morning (10/24/2023 09:41:40 PDT): >who asked you to build 30 tools (10/24/2023 09:41:41 PDT): >nobody (10/24/2023 09:41:46 PDT): >I dunno why we're doing cartwheels to avoid doing the thing we're actually being asked to do *which we can easily do at very little topline cost!* (10/24/2023 09:42:12 PDT): >It's really puzzling IDK what to say. (10/24/2023 09:42:24 PDT): >████ (10/24/2023 09:47:40 PDT): (10/24/2023 09:48:26 PDT): (10/24/2023 09:48:41 PDT): (10/24/2023 09:48:51 PDT): HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-072-00032552 Case 4:22-md-03047-YGR Document 2889-23 Filed 03/27/26 (10/24/2023 09:49:01 PDT): (10/24/2023 09:49:15 PDT): (10/24/2023 09:49:24 PDT): (10/24/2023 09:49:36 PDT): (10/24/2023 09:50:10 PDT): (10/24/2023 09:50:25 PDT): (10/24/2023 09:50:57 PDT): (10/24/2023 09:51:16 PDT): (10/24/2023 09:52:31 PDT): (10/24/2023 11:32:01 PDT): >https://Fb.workplace.com/groups/commsupdates/permalink/25359058740382735/ (10/24/2023 11:32:12 PDT): >i saw. don't get me started. (10/24/2023 11:32:24 PDT): >this is so funny... > >Regulators: Stop optimizing for engagement! >Meta: "We have built 30 tools!" (10/24/2023 11:32:25 PDT): >I'm giving an earful in a meeting now (10/24/2023 11:32:27 PDT): >101 (10/24/2023 11:32:28 PDT): >good (10/24/2023 11:32:53 PDT): >she literally responded to me by saying "but the time management tools" (10/24/2023 11:32:55 PDT): >FFS (10/24/2023 11:34:11 PDT): >I'm gonna need to pull her 1:1 (10/24/2023 11:34:28 PDT): Page 3 of 11 >I think pragmatically what I may do is to shoehorn this into "compliance" saying this is regulatory readiness (10/24/2023 12:14:32 PDT): shared: 395904239_1571448040326703_7256468352372961683_n.png (10/24/2023 12:14:40 PDT): >the rawsuit is extremely explicit about the problem (10/24/2023 12:15:05 PDT): >send me the doc please (10/24/2023 12:15:11 PDT): HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-072-00032553 Case 4:22-md-03047-YGR Document 2889-23 Filed 03/27/26 Page 4 of 11 >i meant to dig it and haven't had time yet (10/24/2023 12:16:10 PDT): shared: California-v-Meta-complaint.pdf (10/24/2023 12:19:43 PDT): >it actually specifically calls out our comms strategy of throwing a bunch of partial mitigations to avoid dealing with the core problem (10/24/2023 12:19:48 PDT): >what a time to be alive (10/24/2023 12:20:20 PDT): >it's a joke. (10/24/2023 12:20:28 PDT): >classic david ginsberg deflection (10/24/2023 12:25:28 PDT): >i really don't like questions like "what do you think is important" asked to me or other CSI leads. >who cares? y'all are doing the work, you tell ME what I should think is important. (10/24/2023 12:25:40 PDT): >Like of course I have opinions but no, i don't want you to optimize for that blindly (10/24/2023 12:25:47 PDT): >I really appreciate you never ask me that (10/24/2023 12:26:05 PDT): >natural function of not having roadmaps (10/24/2023 12:26:21 PDT): >people are so used to being randomly thrown at whatever shiny object of the day exists that it's not worth investing in ideas (10/24/2023 12:39:14 PDT): >oh boy. https://fb.workplace.com/groups/508215109227109/posts/25359235147031761 (10/24/2023 12:39:53 PDT): >sweet comments section (10/24/2023 12:39:59 PDT): >i know. (10/24/2023 12:40:02 PDT): >ooof. (10/24/2023 12:40:11 PDT): >totally doesn't sound like we are biased internally (10/24/2023 12:40:55 PDT): >it is surprising how much of this is redacted (10/24/2023 12:41:29 PDT): >this is so unambiguously about ranking and algorithms (10/24/2023 12:41:35 PDT): >uh (10/24/2023 12:41:51 PDT): >it sure is (10/24/2023 14:16:14 PDT): >https://fb.workplace.com/notes/928815928835484 (10/24/2023 14:16:25 PDT): >I'll give feedback on the writing style here. but i think this is promising (10/24/2023 14:16:50 PDT): >promising for what? (10/24/2023 14:17:04 PDT): >using something like this as guardrail (10/24/2023 14:17:07 PDT): >to not digress? HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-072-00032554 Case 4:22-md-03047-YGR Document 2889-23 Filed 03/27/26 Page 5 of 11 (10/24/2023 14:18:27 PDT): >I dont think we can just turn up to IG relevance with this and say "the current level of RTs is the right level" and block launches that evolve the platform to more unconnected content (10/24/2023 14:19:01 PDT): >ah but i think you could get RT through unconnected content too can't you? (10/24/2023 14:19:06 PDT): >not realпу (10/24/2023 14:19:09 PDT): >the coerricients are also very smol (10/24/2023 14:19:25 PDT): >but you could say, within the connected content you surface, optimize for RT (10/24/2023 14:19:50 PDT): >I would not be confident that is a good idea (10/24/2023 14:19:55 PDT): >it is like, technically possible to make a friend through reels recs, but it doesn't really happen best I can tell (10/24/2023 14:20:20 PDT): >why not? (10/24/2023 14:20:24 PDT): >this just says "people who interact with their friends more on instagram have better social support networks" which like... yes, but so? (10/24/2023 14:21:11 PDT): >yes. and now we have a metric to capture those interactions, and we can see how optimizing for that can be a positive outcome for the platform. >because is all connected content equal? I would say no. (10/24/2023 14:21:36 PDT): >you would just be optimizing for RT inputs, which are like... p(like), p(comment_reply) etc. (10/24/2023 14:21:42 PDT): >So TLS back to MSI based ranking (10/24/2023 14:22:41 PDT): >I think if we are going to make algorithms more principle driven, RT-like metrics are one way to do it. We have sufficient evidence that it correlates with support, which is a good thing. (10/24/2023 14:22:53 PDT): >I dont think so, and the correlations are very weak (10/24/2023 14:23:06 PDT): >Instagram isn't really about RTs, nor should it be. You could plausibly have claimed Facebook was in 2015, but IG in 2023? I don't think anyone thinks that's what it's for (10/24/2023 14:23:25 PDT): >I think there is a whole "communities" org in IG (10/24/2023 14:23:32 PDT): >which tells me they do care a bit about it. (10/24/2023 14:23:46 PDT): >they don't even have a communities surface (10/24/2023 14:23:57 PDT): >FB at least has groups (10/24/2023 14:24:00 PDT): >they don't have to. (10/24/2023 14:24:12 PDT): >they would if they were really focused on it (10/24/2023 14:24:26 PDT): >not if it is not the right strategy. (10/24/2023 14:24:35 PDT): >making a new surface is not always the right solution (10/24/2023 14:24:58 PDT): >if the goal was to make instagram about communities... it would be the right solution HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-072-00032555 Case 4:22-md-03047-YGR Document 2889-23 Filed 03/27/26 Page 6 of 11 (10/24/2023 14:25:18 PDT): >not necessarily! the design space is immense. (10/24/2023 14:25:36 PDT): >████ (10/24/2023 14:25:44 PDT): >████ (10/24/2023 14:25:46 PDT): (10/24/2023 14:25:57 PDT): (10/24/2023 14:26:03 PDT): (10/24/2023 14:26:15 PDT): (10/24/2023 14:26:18 PDT): (10/24/2023 14:26:29 PDT): (10/24/2023 14:26:29 PDT): (10/24/2023 14:26:41 PDT): (10/24/2023 14:26:46 PDT): (10/24/2023 14:26:49 PDT): (10/24/2023 14:26:52 PDT): (10/24/2023 14:27:04 PDT): (10/24/2023 14:27:14 PDT): (10/24/2023 14:27:21 PDT): (10/24/2023 14:27:45 PDT): (10/24/2023 14:27:47 PDT): (10/24/2023 14:27:57 PDT): (10/24/2023 14:28:01 PDT): (10/24/2023 14:28:09 PDT): (10/24/2023 14:28:34 PDT): (10/24/2023 14:34:23 PDT): (10/24/2023 14:34:23 PDT): HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-072-00032556 Case 4:22-md-03047-YGR Document 2889-23 Filed 03/27/26 Page 7 of 11 (10/24/2023 14:35:06 PDT): (10/24/2023 15:03:51 PDT): (10/24/2023 15:04:36 PDT): (10/24/2023 15:04:49 PDT): (10/24/2023 15:05:04 PDT): >████ (10/24/2023 15:05:57 PDT): >████ (10/24/2023 15:06:31 PDT): (10/24/2023 15:06:53 PDT): (10/24/2023 15:07:00 PDT): >does (10/24/2023 15:08:20 PDT): >know that we literally optimize Teen feeds for engagement? (10/24/2023 15:08:41 PDT): >you need to understand that these folks don't know / go into the technical details. (10/24/2023 15:08:44 PDT): -Legal PRIV (10/24/2023 15:08:52 PDT): >that's why I ask, because someone has to tell them (10/24/2023 15:10:52 PDT): PRIV >at the ground floor, they sound like partisan defenders of the company who are wilfully avoiding knowing the things they are refuting (10/24/2023 15:10:53 PDT): shared: 396040808_393873692974528_8150676540219452764_n.png (10/24/2023 15:10:57 PDT): >i had a rather uninspiring conversation with miki about this tbh (10/24/2023 15:11:07 PDT): >just today (10/24/2023 15:11:29 PDT): >he was like "we must have tried optimizing for things other than engagement and it was probably worse" (10/24/2023 15:11:51 PDT): >it was worse for the engagement metrics yes (10/24/2023 15:12:13 PDT): >he had not considered that there may be a universe where we rank content differently for U18 (10/24/2023 15:12:38 PDT): >I think we will need to do something like 'see more clicks per BTPS vpv' (10/24/2023 15:12:43 PDT): >and then started saying things like "but if we are proposing it, i need to know the technical details so i can converse with lars" and my man, that is not the point. are you ok with this on principle. (10/24/2023 15:12:45 PDT): >or Tikes per vpv or something (10/24/2023 15:12:55 PDT): >what do you mean HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-072-00032557 Case 4:22-md-03047-YGR Document 2889-23 Filed 03/27/26 Page 8 of 11 (10/24/2023 15:13:07 PDT): >BTPS = SMSL (10/24/2023 15:13:39 PDT): >SM = CTTck on 'see more' button". But we can't do raw counts of SM clicks without implicitly optimizing for time spent, so it's SM CTR rather than SM count (10/24/2023 15:14:18 PDT): >err does he know that IG ranking is already different for teenagers... and FB ranking is different for YA... just in both cases it's to squeeze even more sessions out of them (10/24/2023 15:25:15 PDT): >I think it'd be a good empathy building exercise to run the FB algorithm on workplace for a little bit (10/24/2023 15:25:29 PDT): >just to see how it elevated binfire posts (10/24/2023 15:40:58 PDT): >I have rocated the workplace algorithm: HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-072-00032558 HIGHLY CONFIDENTIAL (COMPETITOR) HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-072-00032559 Case 4:22-md-03047-YGR Document 2889-23 Filed 03/27/26 Page 10 of 11 META3047MDL-072-00032560 Case 4:22-md-03047-YGR Document 2889-23 Filed 03/27/26 Page 11 of 11 } >} > Case 4:22-md-03047-YGR Document 2889-23 Filed 03/27/26 Page 9 of 11 (10/24/2023 15:59:34 PDT): >what are you doing.... (10/24/2023 15:59:41 PDT): >and omg "blacklist" are you kidding me (10/24/2023 16:00:02 PDT): >I wanted to see what a team optimizing for actual user value might do in a comparable setting (10/24/2023 16:00:19 PDT): >i'm mustering the strength to work in a set up where is repeatedly being asked to put process over my team's work. (10/24/2023 16:00:31 PDT): >fun (10/24/2023 16:00:34 PDT): >Taking all the strength I have to not say "f you all" (10/24/2023 16:00:47 PDT): >maybe we should just give up for a couple years, see if it's better in 2026 (10/24/2023 16:01:12 PDT): >I think you taking parental leave in H2 2024, and me taking parental leave in H1 2025 sound like an amazing idea. (10/24/2023 17:45:41 PDT): >how do they do this every half like clockwork (10/24/2023 17:45:41 PDT): shared: 395412074_285748414366445_6618436928385208323_n.png (10/24/2023 17:46:11 PDT): >it seems to genuinely be the case that optimizing for sessions just naturally distributes more harm to teenagers... it's been too consistent over time (10/24/2023 18:02:57 PDT): >maybe the question we need to ask is, what does compliant teen ranking actually look like. I think it requires: >(a) not being optimized for engagement metrics; >(b) exposing teens to less harm, even if it costs engagement; >(c) niche YIP things > >Can SI tell IG/FB that their approach to teen ranking is non-compliant for failing to meet these criteria? (10/24/2023 18:30:44 PDT): >For (a) we will have to tell them to be compliant what we optimize for (10/24/2023 19:32:34 PDT): >MAP (10/24/2023 19:33:22 PDT): >IG are moving to DAP next year, that's progress HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-072-00032561