This is not an argument against AI. I use it daily, I build with it, and by most measures I'm on the winning side of the system it's creating, which is exactly why this argument needs to be made now, and why I'm in a position to make it.
The debate about artificial intelligence keeps returning to the same question: is it intelligent? I'm not going to spend much time there. LLMs are not intelligent, not because of what they can or can't do, but because of what they are: statistical pattern matching at scale over human-generated data. They're mimicry, not cognition, and that's not a controversial claim if you understand the architecture.
But a lot of people are making a different argument. They're pointing at capability failures: LLMs can't reliably count the letter "r" in "strawberry", therefore they're not intelligent. The problem with that argument is that many humans can't either, because they're illiterate, or because they process language differently. If that's your criterion, you've excluded a significant portion of humanity from intelligence before you've even reached AI, and if you then adjust the criterion to accommodate those humans, you need to explain why the adjusted criterion still excludes LLMs.
