I largely agree with what you said, the problem is phrases like 'simulate hypothetical events' and 'applying this mechanism' are just other ways of saying thinking and do not pin down what thinking is. I know it is not understood what thinking is so I cannot expect you to solve it on Lemmy. I do think it is the manipulation and re-weighting of memory and did therefore feel the original statement that AI models have good memory but poorer ability to think was a little tautologous.

The success of LLM in mimicking human communication makes me wonder if they have unintentionally solved what thinking is and that the human mind is probabilistic. Our minds might decide upon the next event by experience of what successfully followed in similar previously observed situations. This probability work all being run in the subconscious mind, the answer then being passed to the conscious mind which congratulates itself on thinking it out. Similar systems work for picking up a cup, my conscious mind decides I want to pick up a cup but it is the subconscious mind that does all the hard labour of sending appropriate signals to the muscles in my arm. As with reasoning this could have been learned by the subconscious probabilistically. In infancy sending out many signals and seeing which reached for the cup.