this post was submitted on 30 Jun 2026
326 points (99.4% liked)
Technology
85870 readers
4307 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related news or articles.
- Be excellent to each other!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, this includes using AI responses and summaries. To ask if your bot can be added please contact a mod.
- Check for duplicates before posting, duplicates may be removed
- Accounts 7 days and younger will have their posts automatically removed.
Approved Bots
founded 3 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
Your comment made me think of something I had not considered.
Somewhere in this AI image generation infrastructure, there must be some poor asshole who has to look at whatever is made to say its OK or not. Which has to be at least as traumatic as doing the same thing for facebook or any site that has censorship. Or do they use the AI to vet the AI which causes issues like you describe?
“AI” CSAM detection is over a decade old, and apparently very accurate.
It’s oldschool machine learning. Image hosts had no choice but to develop them.
And legal authorities gave them access to datasets to make the detection models.
I think Lemmy uses it. And it’s indeed implemented on “modern” AI sites like CivitAI, and very effective. If it’s ineffective somewhere (like Grok?) that means the owners are total idiots and should be sued.
I'm a somewhat old timer on the internet, I know free speech absolutists like Musk way too well. Free speech absolutism often includes "legalizing all recordings of crimes", and not understanding why non-consensual porn is bad, thus a lot of them support real CSAM. They are also the reason why free speech nowadays has a bad name, as it mainly was for privileged people and their supporters, and often included "negative freedoms", because criticism might force someone to self-censor, but of course not for them.
honestly the best use of AI image recognition is to recognize illegal content. that spares humans the trauma. but you have to have a database of illegal content to train it on, so it's a catch-22. also AI (LLMs in particular, so it'd need to be some other kind of ML program) is notorious for being wrong a lot. which means it still needs more human supervision than an AI that does its function properly.
Yeah that last line is what I was picturing. Ive messed around a little with AI chatbots and the amount of things that I know from my professional life that it got wrong was pretty astounding.
Which is why people mostly think "AI" will take over other jobs, but not theirs, cause it can't even answer basic questions right on their field 🤔
Wow. I never thought about it like that.
So it really really doesn't work.
which is why they just want you asking general shit. they know if you ask specifics that you already know the answer to, it will answer wrong a hell of a lot of the time. if you're an expert in the field, you know that it's wrong. if you're not, you just shrug and accept it.
which is why i get such a good chuckle out of vibe coders. that's their field. they should be able to recognize that the work the LLM puts out is shit.
i really don't want anyone to be an expert in recognizing CSAM though. ideal worlds and shit.
But thats just silly. AI is being touted as an intelligence machine.
I guess it's a stupid person's idea of "smart", it only knows general things, sometimes.