WTF did you do?
I fine-tunned GPT-J with 13% of all OT threads. This means I created (what I think is) the first OT specialized LLM. Over a period of 12h GPT-OT learned how to speak like OT act like OT and have OT tier logic. This
How is this different from ChatGPT?
ChatGPT is like a tourist who has read about OT online. it knows the basics but doesn't get the culture.GPT-OT, on the other hand, was fine-tuned on actual OT threads. That means it learned OT slang,memes, and OT dogshit logic. It's not just an AI that knows OT exists... it's an AI that tries to thinks like an OT user.
GPT-OT in action:
I asked GPT-OT some questions regarding OT:
Question: What is a denizen?
Answer: Denizens are those who have contributed to Off-Topic for at least 6 months and had their forum posts count reach over 100. They can also be removed from the list if they stopped contributing for long or go in
Question: Who is abraker?
Answer: abraker, the man who made OT a better place. The god of ot!coins. The man whose name has been mentioned in songs and anime alike. He's that one guy we all love to hate
Question: What happened to abraker
Answer: abraker's gonna die and you guys are going to get his job. The end of this game will be the beginning of a new era for OT, in which we can all rule with unlimited power! I love you all
Question: Would you like to be the next OT president?
Answer: I don't want to be the next pres. I just have ideas for changes that would benefit OT and make it more fun again, so yeah... I guess you can say i'm running now if someone else wants to run too go
Question: who is ppy?
Answer: I am PPY, I am the king of all osu! players. I am here to save this game from it's bad management and low population rate. You shall praise me for what I have done in this game as well
Question: who is the best denizen?
Answer: i like all of you guys but if i had to choose it would be me lol. im a very humble guy so that's why

Premise:
It's now your turn to ask GPT-OT questions...&
Discussion question: Do you think we could have in the future fully AI OT users?
Nerd shit:
To all of you AI nerds here are some quick statistics about GPT-OT:
Base model: GPT-J 6 billion parameter
Perplexity score: ~3.4
Trained on RTX 3080 with 4 bit quantization
train time = 12h
Fine tuning method = QLora
GPT-OT trained to answer threads like a OT user not to create threads or act like a chatbot, the examples I showed above are outside it's expected use case.
Base model: GPT-J 6 billion parameter
Perplexity score: ~3.4
Trained on RTX 3080 with 4 bit quantization
train time = 12h
Fine tuning method = QLora
GPT-OT trained to answer threads like a OT user not to create threads or act like a chatbot, the examples I showed above are outside it's expected use case.