Meta & Children's Online Safety

Documents / Chat: we probably shouldn't have ranking primarily driven by p(eyeballtime) and p(engagement), which is how it currently works

Chat: we probably shouldn't have ranking primarily driven by p(eyeballtime) and p(engagement), which is how it currently works

Internal document11 messagesParticipants redacted or unnamed

Narratives:Knowledge

Open source PDF (2 pages)

A readable view of the chat, built from the transcribed text. Each row is one timestamped message in the record. Times are Pacific, as recorded. Check the Source PDF for the original layout.

████ name redacted. Different bars may be the same person or different people.

quoted in an excerpt in the archive.

Thursday 10 August 2023

11:44████Great job presenting and fielding questions,
12:44████Thanks all for your work here. Just heard from
12:44████hence the lack of internal alignment that the leads didn't get a chance to do a pre-read
12:44████but hopefully they will meet and discuss in the next couple of days
12:45████Ah! That makes sense based on the reactions they felt novel. And sounds good!
12:53████can they dot vote on where the sliders are... feels like that's where a lot of the meat of the conversation should be
12:54████that's a good idea. I can suggest to
12:54████if any of them feel like "hey actually I think the slider should be x construct not y", great, let's have that discussion
13:27████I worry that none of them are going to be close enough/ invested enough to engage in the conversation thoughtfully/closely enough to get us where we need to go....
14:01████You know i'd rather you tell them than ask them anyway. I think the hierarchy is right, and the principles are largely what is needed to mobilize product change. As mentioned, I think the characterization of "good" and "integrity" only makes sense in naive boosts/demotions, whereas in practice, as you note on slide 5, we have the opportunity to make more fundamental changes for the better that aren't clearly one or the other. I think the piece on more safety-first is uncontroversial and in flight, through lower precision thresholds, age-gating, filtering, demotion etc.. but the more interseting pieces I read from the sub- heading texts are things like: 1. Teens are still developing, novelty-seeking and impulsive... Cool, so we probably shouldn't have ranking primarily driven by p(eyeballtime) and p(engagement), which is how it currently works. We can mixshift towards other terms pretty easily, and this is part of what we're doing with the B&H experiments. 2. Content diversity is important for Teens... Cool, contextual pass spacing rules for FIT topics can be strengthened 3. Teen ranking should be more about EXPLORE than EXPLOIT... Cool, there is a defined mix of these already in prod, so we can mixshift their recs away from optimizing for known interests towards helping them find new ones. etc.. These principles are what is needed on the FI side to justify the work, so I think the most important thing is locking them in such that we can share them and get feedback from surface teams.
14:45████Maybe also that justification of why we need to ratify these internally is important?