>But it’s getting harder and harder to define a task that humans beat LLMs on. On pretty much any easily quantifiable test of knowledge or reasoning, the machines win.

I and likely the person who you replayed to don't find that existing studies actually hold this to be true.