Ep. 525 Gene Callahan Asks: If LLMs Are Thinking, What Are They Thinking About?
Frequent guest Gene Callahan returns to present a talk he’s been giving in academic circles, which relies on standard computer science to argue that LLMs can’t possibly be thinking “about” the external world in the same way humans do. Economist and philosopher Kevin Vallier joins the discussion.
Mentioned in the Episode and Other Links of Interest:
- The YouTube version of this episode.
- Gene Callahan on the ghost in the AI machine.
- Callahan’s homepage. Vallier’s homepage.
- Help support the Bob Murphy Show.

Major problems:
Gene frames the presentation around 101-level intro on computers and algorithms, but offers no primer on the transformer architecture or neural networks. That’s like describing a birthday cake at the atomic level. It’s like describing software at the electronic level.
He’s missing the layer of abstraction that is relevant to the issues at hand. Irrespective of whether the BMS audience is sufficiently equipped to conceptualize what’s going on, it precludes arriving at anything approaching descriptive value.
Gene’s argument that binary 0:1 inputs are inherently meaningless applies directly to neurons firing. We find no philosophical depth there only a waste of time. Then he gets the derivative argument wrong – that neuron firing being binary proves that we can’t ‘know’ about the external world. When we talk about knowing anything, that verb applies to humans, primarily, and always has. Tossing that up is an excercise in postmodern word-salad-ing.
Now that’s at 27 minutes into the presentation and all we can do is click ‘close’. Because our time is finite on this earth.
I pointed out all of these objections to Gene. So here’s what happens when you leave comments about an episode:
A) If the guest says something you don’t like and I let it go, you say I’m serving Satan (literally).
B) If the guest says something you don’t like and I call it out, you ignore what I did and complain about the guest.