Yeah, LLMs cannot learn and that is fundamental to their base algorithms of diffusion and backpropagation. Instead, they can only be trained to asymptotically approach human communication thru the use of latent space, a method of algorithmically abstracting concepts.
They can only ever combine concepts they have been trained on, but cannot continue to build upon those compositions like actual artists do.
Yeah, LLMs cannot learn and that is fundamental to their base algorithms of diffusion and backpropagation. Instead, they can only be trained to asymptotically approach human communication thru the use of latent space, a method of algorithmically abstracting concepts.
They can only ever combine concepts they have been trained on, but cannot continue to build upon those compositions like actual artists do.