ARTICLE

Jared Kaplan

Created 2 May 2025
personresearcherphysicistscaling-lawsanthropicjohns-hopkins

Jared Kaplan

A theoretical physicist who asked a simple question — “does AI performance follow predictable laws, the way physics does?” — and the answer changed the entire field’s strategy.

His 2020 paper proved that AI capability scales as a clean power law with compute, data, and model size. That finding is why companies are spending billions. It’s why the AI race looks the way it does. One paper, from a physics professor at Johns Hopkins University, gave the industry its roadmap.

The Contribution

Before Kaplan’s scaling laws, improving AI meant clever architectural tricks, better datasets, or hope. After it, there was a formula. You could predict how much better a model would be if you doubled its size, or doubled the compute.

That predictability had consequences:

  • It justified massive investment (if you know bigger will work, spending billions is rational)
  • It created the race (whoever scales fastest, wins)
  • It raised the stakes for safety (if capability growth is predictable, you can forecast when dangerous capabilities arrive)

The physics background matters. Kaplan approached neural networks the way a physicist approaches any complex system: measure it, find the laws, understand what governs it. That perspective was novel in ML, where empiricism often stops at “we tried it and it worked.”

The Path

  • Theoretical physics PhD — Studied string theory, quantum gravity, mathematical physics
  • Johns Hopkins University — Assistant Professor of Physics and Astronomy
  • OpenAI collaboration — Worked with the scaling team while at JHU
  • Anthropic co-founder (2021) — Left academia to help build a safety-focused AI lab

The leap from physics to AI isn’t as strange as it sounds. Both fields deal with high-dimensional systems, emergent behaviour, and the question of what fundamental laws govern complex phenomena. Kaplan just pointed that lens at neural networks.

Why He’s Here

Kaplan represents something important about this moment: the insight that AI progress might be law-governed — predictable, measurable, inevitable given enough resources. That’s simultaneously exciting (we can plan) and sobering (we can see what’s coming).

He also made a choice. He saw where the scaling curves pointed and decided to co-found a lab that takes safety seriously at frontier scale. That’s Anthropic‘s core thesis: you have to be at the frontier to study frontier safety. Kaplan’s work is the empirical proof of why the frontier keeps moving.

Key Work

Go Deeper

enes