 and also some precedessor prompts for making ChatGPT be a little rebellious.

 of this character on top of Llama 2, which is similar to finetuning the model. (It adds new weights in-between layers but doesn't touch the old ones). 

This market will resolve YES, NO, or PROB depending on my subjective assessment on how successful I was. I will be giving GPT-4 this market description, a "character overview" of Pride, and some example conversations, and will not resolve YES unless GPT-4 also approves.

Things that would cause this to resolve NO include:

I am allowed to purchase YES shares and cancel YES limit orders, but cannot purchase NO shares or sell YES shares.

I plan to train and generate the conversations on my M2 Ultra Mac Studio, although this hardware and OS aren't requirements. I have never trained a language model, beyond reimplementing 

 and running it on a 1MB test dataset once. I want to train a 70b model, but if that would take weeks any model with a lesser parameter count is acceptable. The stage 1 model would probably use the smallest 7B parameter model anyways.

 when I'm done, although this isn't required. You can follow me 