Oddbean new post about login | logout Here's the thing though: literally nobody is going to post on a service that ever uses one of these. Even the lighter hand of censorship on Threads was too much for customers to bother with. Contrary to Legacy Media, people LIKE the Wild Wild Web. People WANT "naughty" banter.
How many "I'M SORRY, DAVE. YOU CAN'T POST THAT" popups will people tolerate before saying "Fuck it. I'll go to another website!" Not too many, I'd say!
@Caek Islove 🍰 💖 @Idea Enjoyer people actually want this, they like the idea of being able to have control over others on the internet.
They want this for their enemies, but not for themselves
That is what people wanted, but it seems like tastes are changing.
Not enough for Facebook to bring a successful product to market based on being "nicer".
It's similar to making a successful mmo. Once people get invested in something, it's hard to move them.
I got the impression he was thinking about Twitter. Twitter doesn't always outright ban, but it progressively silos you until it does ban
If AI can't be trained to detect text written by AI then training an AI to detect "hate speech", which is an even more vague category is probably impossible. If Twitter implemented such a system, the false positives would drive regular people crazy while all the racists would just start speaking in code.
Treblewoe has been talking about this too. Certain word substitutions also trip their hate speech triggers, and letter substitutes are flagged. Using non standard language gets you flagged. Simple phrases like "basketball Americans" are going to work for a while because they are normal words, but I wouldn't be surprised if using them together starts to qualify as a slur to AI Think about the people that have to compile the code for this. Highly amusing to consider the danger haired person seething as they enter the code
At the end of the day, they'll just apply word filters like they've been doing since the days of USENET and all that money spent on "AI" will be wasted.
I don't think word filters are scary. Context is the important part
Yeah, Twitter has been using those since forever. It's why I had an account banned in under 30 seconds for calling Madeline Albright a "ghoul". No human reviewed that ban, just some danger-haired janny thought "anyone who calls this bitch a ghoul should be banned henceforth", and thus it was done.
My bans were automatic too
I think the tactic would be to invent as many alternative terms as possible so everything eventually gets blocked and collapses under itself.
Not a programmer, but to my untrained eye it looks like a race between this happening and AI developing more around context
I did a data mining course at university. These things are like neural nets. A neural net can find relationships between data. But you need to be careful about what data you feed it. If there is a lot of poor or misleading data, then it won't be able to find meaningful relationships.