>>510450599 (OP)To be fair to that guy, he took an existing GPT model (an older and smaller one) and then trained it on 4chan data. And then ran a kind of experiment with a bot and a VPN to see if it could in fact pass as a 4chan user. Now this guy is a researcher who is very knowledgeable and interested in this field, and he made videos about what he did in a very open way and only ran the experiment for a set amount of time. So I don't think his motives were actually bad here, it was more trying to see if the technology was good enough and if it could work rather than some malicious reason. If he had bad intentions I don't think he'd have talked about it so publicly, in such great detail.
>it's pretty clear the jannies were allowing him to runJannies can only enforce the rules. There are a lot of shitposts here. And it's been that way since I've been on this site on and off since 2008. There were always retards who would write one word replies like "faggot" or "lol". The fact that smartphones make it easier or the internet got browner is beside the point, there was always a lot of noise here and if you were to report the obviously low quality posts, not much would happen. As low effort as it is to just say lol or fag in reply to someone, someone still has the right to say it. In fact, when the board is being spammed with absolute cringe bullshit (like NAFO or shareblue does) the actual cure for it is to call them what they are which are cringe faggots. trannies who will never be real women and so on.
Obviously if it's a bot doing this you can argue that well then it's ineffectual right? Since there's not an actual person behind it. But this place is being used as a training set for the larger models. When they do some work on the dataset and clean it, it will improve the models. Because despite the spam (which is easy for LLMs /deduplication) to clean, there is a great diversity of thought here. So consider refuting LLMs/NPCs as adding to the future training set.
