This rant is part of my reaction to Goolge IO 2026 keynote
What’s the mission for Big LLMTM?
A bicycle for the mindOrganise the world’s informationMake the world more open and connectedSupercomputers everyone’s pockets
The LLM industry seems to be racing to a different mission: all the power of the gods in your pocket. Nothing can’t be known for you. Nothing can’t be imagined for you. Nothing can’t be built for you. With the [new AI app] you have no limits1. You don’t need community, you don’t need friends, you don’t need society, you don’t need to talk to people, you don’t need to learn: you have a god in your pocket.
The web was built on community. On sharing. Forget the web, you don’t need that any more: you have a god in your pocket. We’re post-web now, baby, communities make training data.
(Building gods makes you arrogant, I guess. Or arrogance makes you build gods. Or both.)
Does having personalised pet gods make us selfish? Time will tell is telling.2 But nevermind that, the LLM experience is all about you. Not us. You, the individual you, are very important and powerful. Your pocket-god is all you need (as long as you keep paying).
Every great gift has a curse
As we all know, every gift of god-like powers comes with a curse. Sacrificing your first born. Destructive greed. Flying too close to the sun. Jealous smiting. Read your fairy tales and don’t forget to leave the ball before your tokens run out.
Environmental destruction, of course. The fading of strip-mined digital communities. Isolation flywheels. And intellectual stasis, humanity stuck and barely able to create anything new.
Sometimes our heroes beat the curse. Cinderella married her prince, Aladdin and his genies inherited the throne. And these curses aren’t inevitable outcomes of LLMs. But how Google and others are choosing to build products with LLMs look ominously like a too-good-to-be-true gift, the consequences of which the plucky underdogs will have to overcome before we can live happily ever after.
- As long as someone else knew it before. As long as someone else has imagined it before. As long as someone else has built something similar before. Training data has gotta come from somewhere. ↩︎
- https://www.sciencedirect.com/science/article/pii/S0747563225001566 ↩︎