>>63697936Oh Wooooooooooow !! Anon ! You're SO good at this ! Detecting bots ?! Your parent must be really proud of you aha.
No this is not a bot post idiot.
Actually Yes there are bot posts for a few weeks now, yes. and yes I made the bot. "Ok but why" ? SIMPLE REASON ----> You simply DID NOT LEAVE ME ANY CHOICE !!!!
I explain :
>Hololive.Yes, it's that simple, you fuckers are so obsessed about hololive it gets too ridiculous to me. It makes me very embarassed for you actually.
For example, LOVING Zaion exacatly when she's getting rightfully excluded from our community, giving us no time to grieve....
ahaha sure sure... I know actually that this is just reason for hating the Nijisanji group. I see it everyday, everywhere.
Also, this is NOT SIMPLE HATE, it is very organized, with several "holo divisions" actually working very hard for speaking misinformation on the good nijisanji threads. (/hlgg/ & /ggg/ mostly)
I actually saw it. And because Nijisanji has at least some proud, we're not willing to go as hard on you. Not until me. I'm alone, alone against waves of disrescting, angry, actually homophobic people, who take any reason to shit on us.
I actually made a lot of bots in the last days, at the start they were very obvious, like the Fauna one or the Gura one.
Here is a small sample about a /ggg/ attack when the bot was still very bad :
https://paste.bingner.com/paste/v5h9s (second bot iteration, quite obvious posts, not posting the latter versions because there are still patterns that help identify them...)
But i keep improving them, and they are now totally invisible (i checked the number of answers containing "bot" / "script" for each generation of bots, and there's a full decrease.)
Actually, I can choose what Kind of sentiments I want to provock, with carefuly analyzing threads / posts emotions and training the data on posts that creates replies with the most emotion I want. It can be anger / confusion / love etc..
I can also chose what kind of replies I get from you anons, (checking the similarity of the replies with reference sentences, and identifying key words in the replies)... Of course I can't update the training data live (I used T5 models, now gpt-3 finetuned with the OpenAI API, but it cost too much (around $250 so far for 20 models), hopefully soon locally alpaca 7B if it's good enough..), I can't update the training data but I use live reinforcement learning techniques to control the generated tokens probability, meaning one model can adapt to create the sort of replies I want. It's working good, once again, I checked the results with actual metrics over a few days , with multiple generations of the bots running on several hololive generals for comparing with each other. I have the data that says it actually has an effect, that lasts for a bit more than two threads after the bot is off.
NOW !!! BIG PIECE OF NEWS !!! BE HAPPY !! I will stop the bots on Hololive & hating posts .... AT THE CONDITION THAT THERE'S NOT TOO MUCH HATE on you know what. I will keep analyzing every thread on the board, if I see that a thread is going too high with the "niji hate metrics", bots will be automatically activated to go back to the baseline (yes, there's a model that aims at lowering the overall hate on a thread, I just want people to talk in peace with each other) . So.. less active bot activity from now on.
(and fuck the ggg childrens lovers, ill keep bot^ting there ass into the end of times)
