Toshi Anders Hoo: The Model is the Medium

<aside> 💬

"All models are wrong. Some are useful."

</aside>

Thursday October 29, 2026 · 12:10 PM · Talk

Key ideas from the talk

Futures work aims to widen the cone of possibilities from the probable and plausible out to the preposterous, so people can define preferable futures.
He prefers "future curious" to "future proof". A language model does not change its weights after reading, and curiosity here means the willingness to be changed.
He treats generative AI as a media format and urges model literacy. Ask how a model is wrong and how it is useful, as with any medium.
AI is made of people: the training data, the design decisions and the underpaid workers who tag data and moderate content.
He defines intelligence as modeling a system and updating the model. Pre-trained LLMs do this poorly.
He forecasts a desktop modeling revolution, comparing it to desktop publishing. The main shareable form of collaborative intelligence would move from documents to models.
His own uses include a Claude model of an innovation center that does not exist yet and a chatbot version of himself loaded with his talk data.

"No one can predict the future. Don't believe anybody that tells you they can predict the future."
"If you think you know what's going to happen, you're most likely going to miss what's actually happening."
"Our brains are wired to think about everything that's happened to us in the past, create a mental model of the world, and we project that into the present and also into the future."
"AI is made of people. It's made of humans."
"I think we're living in the age of models and model making and model sharing."
"I believe that we're on the cusp of a desktop modeling revolution."
"Language models are not the world; they're a model of the description of the world."