I believe that people who claim that AI is unable to think are not underestimating AI so much as overestimating human ability to think

Maybe, but for example, even a dumb human understands that people have five fingers but AI can’t seem to stop making images of people with six fingers or three arms because it doesn’t understand what the image is on a conceptual level. Additionally, human cognition isn’t sentence prediction which is fundamentally what LLMs are doing.

Now this may require a rethinking of what “thinking” is but I think AI is doing something different, and people aren’t understanding that (and are thus overestimating or underestimating it depending on the circumstance).