Facebook's LLAMA, which is like GPT 3.5 (which ChatGPT is a fine-tune of) uses 4x less memory than GPT3.5 and outperforms its generation. Running in 8bit you can fit the 13 billion parameter model (there's also a 65 billion parameter model, which would perform better, but would need to be running on a commercial GPU with 80GB of video RAM) into 24GB of VRAM, which means a GTX 3090 or 4090 will do it.

Initially the models weren't released and you had to request access from Faceberg, and they were only giving it out to .edu addresses. Someone leaked it into a torrent and now it's everywhere, though.

The cool thing about this? You can train it to eradicate the faggot shitlib bias and also fine-tune it on e.g., 4Chan, Poal, Voat etc. or whatever else you wanted to. The "default" even will be far less censored than e.g., ChatGPT.

And of course it's free to use.

Eh he's right. (((Sam Altman))) and other kikes involved in major AI companies are going to use them mostly for dogshit purposes while neutering the tech to the point it's next to worthless for anyone that isn't in their kike globalist circle to use.

That said I don't see this as too big a problem because MOST of this tech is open sourced, it just takes someone willing to spend the time to figure out how to run, train, prepare data sets etc and pay the bills for running it on cloud machines.