ChatGPT can discuss chess brilliantly but routinely makes illegal moves during actual gameplay. AI researcher Gary Marcus argues this isn’t a quirk—it reveals that large language models lack “world models,” the internal representations of reality that humans use to understand how things work. This limitation may be fundamental to current AI architectures, with significant implications for how we should think about and use these powerful but imperfect tools.
Category: Musings
The Freedom to Speak: An Interview with John Stuart Mill
Nearly 20 years ago, I tackled one of democracy's hardest questions by "interviewing" John Stuart Mill. Here's what he told me about free speech.
What I Learned When Dad Was Dying
Paying attention to what matters most is harder than ever—and just as important as it always has been.
Sometimes the Deep Magic Prevails
Reflections from a wedding on a Blue Ridge mountainside.
What AI Will Never Know About Ice Cream, June 1981, and Everything Before or Since
Despite regular claims to the contrary, AI Does Not Have Access to All of Human Knowledge — and It Never Will
Here’s to the Householders
The people who hold together households hold together society—and we should treat them as such.
Seeing Is Not Believing
We see a great deal that we fail to notice and which we therefore never believe.
Check Your Lenses
Different tools enable different types of seeing and thinking.
When You Catch Yourself Looking Down on Others, Look Up
It’s only human to look down on others. But it’s hard to move onward or upward when you’re looking downward.
On Rewriting History
It's crucial to distinguish between the actual facts of the past and the stories we tell about them in the present. It’s also crucial to recognize that both are driving forces of the future.
