I think this is my main point- isn't it amazing that a thing that predicts words other humans have previously written manages to appear intelligent, or, more pointedly, have utility in communicating real thoughts and ideas?
If you've ever asked an LLM a question and gotten a satisfying answer, that means that there is some human-level intelligence somewhere in the token filtering / recombinating that an LLM does.
Specifically I think the test of human-like intelligence is literally the output- If we get utility from the arrangements of the tokens it outputs, that in and of itself demonstrates that some portion of human intelligence could be this same token generation mechanic.
It is like sitting down at a piano, sight reading a piece from sheet music and then someone who has no idea what they are talking about claiming you composed the music on the fly. Then when you point out the sheet music they just double down on some bullshit as to why they are still right and that is still composing even though obviously it is not.
Can you list a few demonstrations from a text-outputting computer program that would each convince you that there is intelligence here? Eg writing a bestselling novel, proving an unsolved number theory conjecture, etc. Or is your belief uncontestable?