This is a reason why, before LLMs truly become contenders for replacing humans, the sharp distinction between pre-training, fine-tuning, and providing context i

This is a reason why, before LLMs truly become contenders for replacing humans, the sharp distinction between pre-training, fine-tuning, and providing context in a conversation has to disappear. The model must be able to update itself (learn!) by calibrating all these massively multidimensional parameters on the fly, as they operate, like people do.