An LLM could be thinking in one of two ways. Either between adding each individual token, or collectively across multiple tokens. At the individual token level the physical mechanism doesn’t seem to fit the definition being essentially reflexive action, but across multiple tokens that’s a little more questionable especially as multiple approaches are used.
> across multiple tokens
- but how many ? how many of them happen in sole person life ? How many in some calculation ? Does it matter, if a calculation doesn't reflect it but stay all the same ? (conversation with.. a radio - would it have any sense ?)
The connotation is simply an internal process of indeterminate length rather than one of reflexive length. So they don’t apply it when a GPU is slinging out 120 FPS in a first person shooter.
But now, I see this: the truth is static and non-profit, but calculating something can be sold again and again, if you have a hammer (processing) everything looks like a nail, to sell well the word thinking had to be used instead of excuse for every time results being different (like the shadows) - then, we can have only things that let someone else keep making profits: JS, LLM, whatever.. (just not.. "XSLT" alike).
(yet, I need to study for your second sentence;)
(In math class I've been told that not some number but the formula is the expected result..
.. then people are now asking: "Just show me the prompt instead !" and.. done. ;) )