Summary
Does artificial intelligence actually understand anything? David Shapiro takes on a viral post by tech founder Gregory Kurtzer, the co-founder of CentOS and founder of Rocky Linux, who argued that large language models are only statistical token predictors with no real understanding.
Shapiro answers with two frameworks. First, he applies Bloom’s taxonomy, the educational ladder of remember, understand, apply, analyze, evaluate, and create, and points to AI results at each rung: Humanity’s Last Exam scores, Harvard’s AI tutoring study, Google DeepMind’s gold-medal score at the International Mathematical Olympiad, DARPA’s AI Cyber Challenge, GPT-4 as a grader, AlphaEvolve’s matrix multiplication record, and a team of Claude agents writing a C compiler that can build the Linux kernel.
Second, he invokes Ashby’s law of requisite variety and the Conant-Ashby theorem, which holds that every good regulator must be a model of the system it regulates. He cites interpretability work on an Othello-playing model that built an internal board representation.
The video argues that the disagreement ultimately comes from confusing understanding with consciousness. It draws on Alan Turing’s 1950 reply and the illusion of explanatory depth to make the case that understanding should be judged by what a system can do.
📺 Source: David Shapiro · Published October 28, 2026
🏷️ Format: Opinion Editorial
