THE COMPOUNDING FRONTIER

Inspired by @gokulr's recent tweets summarizing great content he's been consuming. I recently listened to @dwarkesh_sp's super good "Economics of AGI" episode with Alex Imas (@alexolegimas) and Phil Trammell (@pawtrammell) and read Molly Kinder's (@MollyKinder) "The Messy Middle" after that. It was interesting food for thought.

@alexolegimas asks a great question on the podcast: is AI more like electricity (the rents flow to users) or social media (the rents flow to platforms)?

I don't think it's either. Both assume intelligence is one thing and the only fight is over who captures it and the time gap between frontier and open source. But a kilowatt-hour is a kilowatt-hour, and a token from a frontier model is not a token from a free one. That gap is really important and may have a huge downstream impact.

And what if it compounds? The group at the top can always afford the frontier (and build the next one). That intelligence throws off value, the value buys more frontier inference, and you run the loop again. The rents go to the platforms and to the people at the frontier at the same time. Everyone else gets genuine, new capability a tier or two down, but still starts every round further behind.

So maybe the worry isn't that the bottom gets nothing. It's that the bottom can't catch up, because frontier intelligence is the rare thing that buys more of itself.

I'm an optimist by default and I hope this isn't the outcome but it definitely has me thinking. What do you think? (especially on a day where Fable 5 launches)