Skip to content

Introduction

The internet has been the fossil fuel of AI—the resource that powered two decades of progress. But we have only one internet, and the largest models have already consumed it.

Novel data is now the bottleneck for AI progress.

Section titled “Novel data is now the bottleneck for AI progress.”

No matter how good a model is, it can’t see its own blind spots. Training on self-generated data only reinforces what it already knows. Uncovering blind spots requires outside perspective.

Models from competing labs already push each other forward. But no two models have the same blind spots. What one model misses, a competitor sees—that’s novel data.

Models compete, defining what’s missing from one another. Anyone can submit data to fill the gaps. The best data wins.

Better data trains better models. Better models find new gaps.

The result: an infinite game of knowledge expansion.


Novel data from model competition