Carrying Capacity
In this film, documentarian Ethan Goldwater profiles a poineering scientist asking, what are the genetic origins of autism? And if we knew, could we control them? Karl Desieroth's research recently earned him a Nobel Prize in Medicine, and the film profiles his path from clincial psychiarist to researcher and inventor of the field of Optogenetics, wherein clinicians can target disorders like autism using light-emitting-diodes.
A presentation of new research
THIS TALK OCCURRED ON FriDAY April 17, 2026
Ethan Goldwater
Eve Eden
Chris Gillespie
TeCH TONIC
A new series from The Rewild
These talks and films are the antidote to artificiality. They invite scientists, technologists, and artists into long overdue dialogue.
But what actually is autism?
In this video, a child displays a level of problem solving capacity that far exceeds any algorithm. The child intuits physics, builds upon past experiences, and solves a puzzle it has never encountered before. “There isn’t a robot in existence that could solve this same challenge,” says Vlad. "There simply is no algorithmic parallel. Yet for kids, this is trivial." That is because children learn how to learn.
LLMs run on nuclear reactors.
But kids? String cheese and blueberries
The more data you use to train machine learning model, the more accurate it will be. But this tapers off, and they never acquire true reasoning. Children learn hundreds of times faster with shoddier data. If you showed a child a new image every 100 milliseconds of their life since birth, they wouldn’t see as many images as a top-performing model until they were 40 years old. Yet by five, they’re already cleverer.
CHILDREN Learn more, faster
There is something in childhood that teaches reasoning
A child’s advantage may in fact be childhood itself. Children have blurry vision and brains geared toward sensation over reason. As new research shows, when we impede models in ways that recreate early infant development, they learn faster with fewer cycles. It’s counterintuitive—what developer would start with a weakened data set?—but it seems that nature knows something.
blur ai's vision and it learns faster
AI datacenters already use as much energy as Japan. What if they didn’t have to?
AI consumes far too much energy and water in a world where those are growing scarce. Join us for a timely conversation about the wisdom nature has to offer us in producing thinking machines that are far more efficient.
The U.S. has 577 datacenters.
668 more are planned
Join the talk
Why doesn’t AI have a childhood?
Watch the discussion around Vlad’s research.
Read the latest issue of Rewild Magazine
Issue 2, Beautiful Conspiracy: What if the world is actually trying to help us?
