>>96021698
>The calculation methodology doesn't preclude "really thinking"
I meant it as it doesn't form abstract rules independent of language: it just searches for the most likely next token to slap down given the history of tokens in its current statement/chat. The AI for a grunt in a vidya or a chess playing bot doesn't "really think" if you want to get all metaphysical, but it does think in that it has a series of rules based on the task at hand.

Instead of going "I should play Card X to defeat Card Y because Rule 7 says that will happen", it just goes "After doing enough random shit, playing Card X after I see Card Y tends to end in a win". It doesn't have actual planning or objectives the same way something like a chess bot would actually be "thinking" some number of moves ahead or a shooting game enemy "thinks" "I am being attacked; cover protects me; I should move to the nearest 'cover' node on the map".

Go look at that Claude Plays Pokemon thing if you want an example of the sort of "thinking" LLMs do.