Why benchmarkable skills are less valued, and why being unbenchmarkable is more necessary in the AI Agent Era.

How next-token prediction and score matching turn intractable generation into tractable, atomic tasks—and why the resulting optimization landscape makes modern AIGC both possible and elegant.

A philosophical essay on meaning as use, normativity, and judgment—and on what contemporary large language models reveal about the power and limits of linguistic fluency.

I predict AI won't converge to a single unifying theory, but rather to a rich tapestry of frameworks—each illuminating different facets of learning, reasoning, and intelligence.