@alberand I plan to make future attempts at automuse read a lot better, though that may end up being done in private repos until I publish another "paper" about the changes being made. I really wanna try using GPT-4 for this, but I also need to consider the costs involved.

@cadey btw do you have any ideas on how to make it context aware? I was playing with llama and making it context aware would bring so many possibilities.

@alberand What do you mean by that?

@cadey in your article you describe it as context window

@alberand Ah, what I'm referring there is more of maintaining a "window" of prior writing blurb context so that it can attempt to continue the story where it left off. The "window" would be implemented as an array of tokens that would be pushed to and popped from so that the total amount of context shared between each run of the AI (prompt and prior context) doesn't breach about 1000 tokens (mostly so that I can maximize the consistency of the output of the AI and minimize costs). My hope is that a model like gpt-4-32k will give me better ways to handle the context window size (so that there can be more story context each run, making things more internally consistent).