October 2024 marked a critical inflection point in AI development. Hidden in the performance data, a subtle elbow emerged - a mathematical harbinger that would prove prophetic.

What began as a minor statistical anomaly has since exploded into exponential growth.

Since then AI performance has surged attaining a new trajectory, a new slope - no longer linear but geometric.

Segmenting out the models by size and type, the source of the innovation is clear. No longer model size which drove the initial wave of improvements, nor the improvements in the smaller models of the early fall.

It’s reasoning - ask a model to articulate its thought process, consider alternatives, & ultimatey select one.

With improved accuracy, fewer errors, & the ability to conduct deep research - work extending for fifteen minutes or more, the potential of the technology has never felt more tangible.

Recently, Alberto Romero suggested that the differences between the performance of AI models is much less important than the difference between people’s ability to use them well. A sophisticated user of AI - like any skilled worker - can produce much more than a novice.

As these models continue to improve, it may be less important for management teams to track relative benchmarks of AI performance & much more to train their teams & reimagine their workflows.