lol

Think of what you're saying, they'll put us out of a job!

why use expensive computation time when you can get humans to do it for free
but on a serious note so much of 4chan's moderation is contextual, and part of the fun is that it's not perfect
one place that it COULD be useful for is detection NWS on SFW boards, as detection for that is pretty good now

>>10342
yeah the idea's not auto-moderation, but to craft a prompt that would rate how bad a post might be and add it to the queue for human review

chat gpt can check my dubs

I don't know anything about the actual numbers involved but I have to assume it would be more costly than you'd think considering the volume of posts here

>>10339
no, fuck AI.

>>10345
I assume this too. Basic screening via bad word and bad phrase lists is cheap, but throwing EVERY post at an LLM is the kind of "burn money to get an audience" loss leader that people do when they've got VC money and then panic when it doesn't make them dominate the whole market within 18 months.

Especially if all it's doing is reporting things that would otherwise go unreported. That limits it to
>Things that no user is bothered about
>Things that users are specifically trying to keep hidden

Jannies and users do a good job of detecting and reporting GR1-related discussion even when users try to keep it subtle.
And if an autojanny is flooding the flesh-jannies' queues with things that might be breaking less important rules and that no user is bothered enough to report, that's going to make more work for the janiteam, which means hiring more of us to cope with the output of the LLM janny.

Yes, there are some threads where users see rulebreaking shit, don't report it, and then post "OMG why is janny not removing this, the porn has been up for 5 hours", but it is still cheaper and easier for us to try to convince users to report blatant rulebreaking than it would be for 4chan to implement a full blown AI report queue.

Not to mention, things that aren't being reported and could in theory be removed by stricter moderation and aren't bothering users - that's the kind of "freedom to argue, shitpost, and be merry" that 4chan has as a unique selling point.

The only problem that I think that ML (and not a LLM btw) would be good at solving is false reports:
>user makes a report on a high clear rate board
>make a copy of said report with the user's IP and add it to a list
>look in the list for users that have more than X reports
>feed the list to an ML model written in Python to categorize individual reports as true or false
>use Python analyze the results
>send the worthwhile users' reports to a false report queue sorted by IPs
>a Mods can then look at the false report queue when they feel like it to warn/ban the actual false reporters
This can:
>cost as little or as much as you want (mostly electricity costs from running a GPU)
>work on as many or as few boards as you need it to (the more reports you have to go through and the larger the training data the longer it will take to run)
You get a system:
>that can find false reporters without human input
>which isn't making any decisions for humans
>that streamlines the handling of false reports
>lowers false reports over time
You'd need:
>some python
>some ML knowledge
>a GPU
>a server to strap said GPU into
This should be able to work pretty well but the most critical and time consuming part would be training the model so that it's accurate. Even if you end up with a model that can categorize stuff correctly 80% of the time the actual accuracy of the entire system "should" be a higher. You'd still need some willing test subjects to judge the actual accuracy of the queue.

>>10339
ur tryna take my fuckin job away no fuck u