> Eventually people will start adopting the distribution used by LLMs, making classification harder.

I recently heard someone say "that's genuinely the exact solution I was looking for" and had to do a double take.