> LLMs are powerful autocomplete with vast and broad memory, they do not think.

This is really only true insofar as they are designed such that they output language tokens and their intermediate "chain of thought" is also done in language tokens, but now that's really only done for auditability. There are much more abstract layers before the model output layer of a single token prediciton and they could be designed to to "think" using these more abstract layers as intermediates instead like how many people think in concepts rather than words. There are even some research models on this (see Next Latent Models)[1]. Humans need to communicate in language because we don't have telepathic abilities but these models need not have such a constraint.

[1] https://arxiv.org/abs/2511.05963