Post B6nXK8iLerymcVh68W by wolf480pl@mstdn.io
(DIR) More posts by wolf480pl@mstdn.io
(DIR) Post #B6nP9ORM3AMOZg6tjk by david@tech.lgbt
1 likes, 0 repeats
"What percentage of social media users do you think post severely toxic content?"https://thenoisyroom.com/
(DIR) Post #B6nPDeMWnHMb79slrk by quad@akko.quad.moe
0 likes, 0 repeats
@david huh
(DIR) Post #B6nPf1SZ1hWGZIl3NQ by david@tech.lgbt
1 likes, 0 repeats
@quad I also guessed 5%!
(DIR) Post #B6nPf1ecItA1AgOgAC by quad@akko.quad.moe
0 likes, 0 repeats
@david i dunno, i don't think i'm that calibrated, my bar for what counts as "severely toxic" is probably just stricter
(DIR) Post #B6nPrux0XPci9699G4 by quad@akko.quad.moe
0 likes, 0 repeats
@david nevermind, apparently i am decently tuned
(DIR) Post #B6nQVFXTvPPrIDxhaa by quad@akko.quad.moe
0 likes, 0 repeats
Dunno if I support the argument at the end to add opinion polls to everything, seems to easy too bot just like likes or boosts.But otherwise this website is a decent readRE: https://tech.lgbt/@david/116659006573176516
(DIR) Post #B6nQmmzkn0uJygY84O by quad@akko.quad.moe
0 likes, 0 repeats
Like they try to explain it away with stratified random sampling. But I don't see how that works online.It works in real life because it's pretty difficult to just invent new birth certificates to increase your chance of being randomly selected. But online I don't think this helps at all because stuff like making a crapload of non-posting accounts with fake identities to read random content and skew polls as they're picked is something you can... just do, like pretty easily.
(DIR) Post #B6nRaW9xenwAF82JVY by kopper@not-brain.d.on-t.work
0 likes, 0 repeats
@david gonna be honest this feels unnervingly LLM generated. i'd appreciate if someone verifies the sources but sentences like "The topic-matching confidence threshold is 0.8" (0.8 of WHAT, exactly?) does not really inspire confidence
(DIR) Post #B6nRaWSkWwxxBOpJnE by natty@astolfo.social
0 likes, 0 repeats
@kopper@not-brain.d.on-t.work @david@tech.lgbt tldr and can't Ctrl-F for itBut confidence threshold can be a unitless quantity just fine
(DIR) Post #B6nSLXDHYddT7Z0ct6 by kopper@not-brain.d.on-t.work
1 likes, 0 repeats
@natty @david it's in one of the dropdowns at the end, ctrl+f would've worked if it was a details but that's what we get when LLMs get involved
(DIR) Post #B6nXK8iLerymcVh68W by wolf480pl@mstdn.io
0 likes, 0 repeats
@quad I think this part:> Coordinated response patterns are anomaly-detected and excludedis supposed to deal with that, but I have no clue how that works
(DIR) Post #B6nZ8PjmjyEVqku2Iy by quad@akko.quad.moe
0 likes, 0 repeats
@wolf480pl Well yeah, I'm well aware that that's what they were going for.But they basically just said "we'll block bots" with zero example of how the heck they'd manage to do that.We've been trying to block bots and automated attacks ever since the first captcha came along (and realistically before that as well), if we still haven't figured it out, i don't see how these people would have any better chance at doing so.
(DIR) Post #B6nZMMl6c0G0fLAnxI by wolf480pl@mstdn.io
0 likes, 0 repeats
@quad no, they didn't say they will block botsthey said they will filter bot votes after the bots have already votedI guess this is kinda like figuring out which websites the spam bots link to, and then finding all posts on your forum that link to those sites and mass-deleting thembut I don't know they're gonna have enough information in a vote to be able to discern that...
(DIR) Post #B6nZhF2uU1N4pHSIMK by quad@akko.quad.moe
0 likes, 0 repeats
@wolf480pl >they said they will filter bot votes after the bots have already votedeh, same shit different ass>but I don't know they're gonna have enough information in a vote to be able to discern that...Even with more info, especially in the age of LLMs that's especially difficult. Even Steam can't hope to figure it out, they add a notice telling you that a game might've been review bombed, but that's mostly it.If from a computer's perspective it basically only sees "1, 2, 3 or 4", how the heck is it supposed to be able to tell if a flood of 3s is made of illegitimate votes or if it's just a legit popular opinion? Are they supposed to supplement it with a lot of device info and fingerprinting which they then cram into an algorithm to guess if you're human or not? At that point it loops back to "no thanks". And even if they did it'd end up like the captcha system, where anyone running linux or another "less common" configuration would essentially be excluded from the vote.
(DIR) Post #B6naGBsuEFxj9mRP7Y by quad@akko.quad.moe
0 likes, 0 repeats
@wolf480pl Even if we make the REALLY generous assumption that they manage to perfectly discern a bot vote. How do you discern a common opinion and a "coordinated response" if they're both from humans.The algorithm needs to somehow know if all the "3" responses it gets are genuinely what people believe, or if people just clicked it to be funny.Even without bots, the internet WILL constantly try to make another "Boaty McBoatface" happen. Without needing to coordinate.