I am trying something new here on Substack.

I’ve been working on an essay about understanding, provisionally and rather uncreatively titled Understanding Understanding.

The premise is simple enough: LLMs are routinely described as subhuman because, among other alleged deficiencies, they lack consciousness and don’t understand.

I don’t argue that LLMs are conscious, nor that they understand. My argument is prior to either claim: understanding is radically underspecified. We use the term to designate a jumble of different capacities, accomplishments, dispositions, experiences, and metaphysical suppositions, none of which appears to be both clearly defined and uniquely human. This is the problem the essay attempts to expose.

The experiment, and the something new, is that I’m sharing a Google Notebook summary generated from my annotated outline. It’s not the essay, nor a substitute for it, but a provisional sign of intent and orientation: an indication of the argument’s present shape before the prose hardens around it.

As usual, I am suspicious of language. Many apparent disagreements are sustained by obscurity, and many supposedly decisive claims conceal their weakest premises inside familiar words.

As always, commentary, advice, objections, and dissent are welcome.

PS

A total aside: I typed this note and ran the ‘Scan for AI text‘ feature. I tells me that this is likely 100% AI, which is to say 0% Human. To add insult to injury, I amended this postscript and reran the process, which inverted the score.

I’m sorry, but this is very unreliable. Anyway, I hope you find the podcast summary engaging. Cheers.