[throw-the-towel]

But can we say that the LLMs are consciously designed? AFAIK nobody really understands why they work.

[amoss]

Whether they are or not is irrelevant to the difference- there is still an explicit division between training and inference, with reinforcement of the statistics of the training set.