Comment by brcmthrowaway

Why is this the case?

Are there any architectures that don't rely on feeding the entire history back into the chat?

Recurrent LLMs?