it is literally impossible to do this at scale without machine learning. hundreds of hysterical morons shouting about something which has been essential to every abuse system for more than a decade. - ThreadSky

theophite.bsky.social • 16 days ago

it is literally impossible to do this at scale without machine learning. hundreds of hysterical morons shouting about something which has been essential to every abuse system for more than a decade.

Reposted from ROOST

AND WE'RE HATCHED!

ROOST launches at the #ParisAIActionSummit, bringing together tech companies and philanthropies to deliver free, open-source safety tools. Join us 👉 roost.tools

Comments

jauntywk.bsky.social•16 days ago

I'm a little afraid of turning the tables such that sites become more liable because there's an expectation that they run one of the big AI moderation systems.

pettter.bsky.social•16 days ago

Have 'foundation models' been essential for more than a decade? Really?

rllyniceguy.bsky.social•15 days ago

all this stuff is one long continuous process that's been developing for fifty+ years. people don't seem to understand that this isn't a new field. it's just hit an inflection point. foundation models are deep learning. machine learning is AI. it's all one umbrella.

pettter.bsky.social•15 days ago

When you apply for money, it's AI, when you hire, it's ML, when you develop, it's logistic/linear regression.

theophite.bsky.social•16 days ago

i don't know that that question is sufficiently coherent to answer

theophite.bsky.social•16 days ago

the short answer is 'yes.' the long answer is 'the definition of that term which you believe in is not what it means.'

pettter.bsky.social•16 days ago

What ROOTS mean by 'foundation models' is LLMs. That word filters were used over a decade ago is not in question. That there were ML techniques used for content moderation over a decade ago is not in question.

That LLMs were _essential_ for content moderation at that time is very much in question.

forthrast.bsky.social•16 days ago

awesome that the plurality of this site is stupid, better than evil but maybe more annoying

jeanpauldifranco.bsky.social•16 days ago

Forced Gen AI basically radicalized everyone against any type of machine learning and it’s a shame. I get why people don’t trust it, but like you said, it’s the only way to moderate at scale

conjurial.bsky.social•16 days ago

The only even vaguely workable alternative is to have users do moderation and to add enough friction that the moderation load is manageable. In other words, a mailing list

And I think usenet shows that didn’t work well either

raspberrycordial.bsky.social•16 days ago

The alternative is to outsource the task to exploitive cheap labor call centers in poor countries, which I'm sure they would also have issues with that

mikeyass.bsky.social•16 days ago

well the alternative is no social media, which might be ok?

uckema.com•15 days ago

No the legit alternative is no images or video on the internet period return to BBs (and even then you have ASCII art!)

mikeyass.bsky.social•15 days ago

what? dude every forum poster for years had trusted uploads for members and mods that scanned the shit.

you just pay to host your own shit. and you own the legal responsibility.

uckema.com•15 days ago

You also had to pay 10 bucks and SA literally topped out at like 10K people tops. Blue sky in like 3 months of being public is literally 100 times the size of peak SA. It’s why 4chan has to use AI bots for mods

mikeyass.bsky.social•15 days ago

i paid my :tenbux:. i still post on forums. it is a better experience than the apps. the conversation is limited to about 100-200 people, and thats fundamentally _better_

group chats are the future

uckema.com•15 days ago

WhatsApp Discord and like every chat program has AI bots searching chats for illegal material by the way

uckema.com•16 days ago

the only other alternative at scale would be Lena style uploads and frying them out every 10 (subjective time) days

publicaardvark.bsky.social•16 days ago

I hate the extent to which stupid 'AI' boosterism, tasteless or malicious use of generative models, and the well-earned anti-SV backlash has made it impossible to have any kind of knowledgeable or rational conversation about existing applications of ML or all the interesting stuff these models do

chathamharrison.bsky.social•15 days ago

Seems to me that to the extent ML is good, it is kinda boring, & seems like it would mostly replace really boring work (not to be rude as I speak in relative ignorance.) Which is good!—but not sexy, & so much trouble seems to have come from trying to make it sexy.

ohdeerthermos.bsky.social•16 days ago

I feel like we're reaching a tipping point where "AI" has poisoned the zeitgeist so much that AI needs a rebrand

wormbly.bsky.social•15 days ago

We, the people, demand that an employee at Google personally checks every indexed image for CSAM

vector-of-bool.bsky.social•16 days ago

Wow, I think I just quintupled the size of my blocked accounts list just while scrolling through that.

toomanyfembots.bsky.social•16 days ago

Unfortunately, this is what happens when AI and ML become coded as right-wing. Turns into a tripwire. I do worry about the consequences of progressives removing themselves from the tech space…

beneprism.rip•16 days ago

Progressives aren't removing themselves from tech, really, but progressives who were already outside of tech have gotten more hostile (justifiably and otherwise)

rainsurname.bsky.social•16 days ago

I understand machine learning is necessary for moderation at scale, but I have always felt that every highly paid person at these companies should have to spend a few hours moderating every month, just so it isn't such an abstract problem.

twitteraddict.bsky.social•16 days ago

Just what we need, for the people running Silicon Valley companies to be even more messed up

rainsurname.bsky.social•16 days ago

I am not 100% serious, but if an affluent person with access to excellent health care couldn't manage four hours a month without being damaged by it, that makes outsourcing it to poorly paid contractors who get no support an obscenity.

walmsley.bsky.social•16 days ago

don't feel great about the choice people are making between "learning one (1) fact about machine learning" and "taking a hammer to load-bearing internet infrastructure because you hate LLMs"

vala.wtf•16 days ago

if/when we get out of the current situation, the tech backlash is going to be so painful

maxp94.bsky.social•15 days ago

I’m fine with this, but mainly because I think the internet is basically bad.

showtooltip.bsky.social•16 days ago

none of them have hammers, it’s fine

Comments

Posting Rules

Reply