Frontier Debates: Language, Vision, and the Boundaries of the World

"Silicon Valley has been hypnotized by LLMs, completely hypnotized by them."

-- Xie Saining

This is a technical judgment: a judgment about a fundamental dispute over two competing definitions of "intelligence," a dispute that is shaping the direction of AI research for the next decade.

This lecture gives no verdict. Four debates, each laying out the sharpest arguments on every side, with the questions left for you to answer.

• Debates 1 and 2 (see "Language as Opium and the Bitter Lesson"): Is language a tool, or "opium"? What did the Bitter Lesson actually say, and have LLMs followed it or violated it?
• Debate 3 (see "Division of Labor and Convergence between World Models and LLMs"): Are world models and LLMs competitors or complements? What is WAM betting on, and what is JEPA betting on? Will the two paths eventually converge?
• Debate 4 (see "Where Does the Data Come From"): In the era of "downloading humanity," where does physical-world data come from?
• Architecture bets and closing questions (see "Core Bets by Architecture Line and Closing Questions"): The core bets of six architecture lines, the CWM case study, and three questions to carry forward.