New releaseThe Anthropic Book ↗

19 chapters  ·  corpus-verified  ·  audio narration

CONTEXT JAMMING

Field notes from inside the context window.

Hire me

The Holographic Universe · one connection

Why is there a line between Enrico Fermi and Jared Kaplan?

idea lineageRests on: shared techniqueEnrico Fermi · Jared Kaplan · 2020

The claim

They never overlapped: Fermi died in 1954, and Kaplan finished his PhD in 2009. The line records a method. In a lecture in January 2020, Kaplan estimated the brain’s computing rate from three round numbers, multiplied by a lifetime, and compared the result with the largest training runs of the day. Physicists call that kind of calculation a Fermi estimate. Kaplan did not.

Why it matters

It is the most honest picture on the map of what physicists carried into AI labs. It was not the content of holography. It was a habit: get the order of magnitude first, then decide whether the question is interesting. Kaplan’s estimate put 2020’s largest training runs within a factor of ten to a hundred of a human lifetime of brain activity. The lecture was posted on January 5, 2020, eighteen days before his scaling-laws paper reached arXiv.

Where to check it

  1. Jared Kaplan, Machine Learning II, Israel Institute for Advanced Studies (video, from 55:40)The estimate, in his words: about 10¹¹ neurons, times about 10⁴ synapses each, times roughly one firing a second, gives about 10¹⁵ operations a second, a petaflop. Over a hundred years that is about 3 × 10²⁴ operations, which he sets beside the largest training runs. He calls his training-cost rule a “guesstimate.”
  2. arXiv:2103.05784, J. I. Katz, Fermi at TrinityWhere the method’s name comes from. At the Trinity test in 1945 Fermi estimated the yield at about 10 kilotons by dropping scraps of paper in the blast wave. Katz notes that no account survives of how he did the arithmetic.
  3. arXiv:2001.08361, Scaling Laws for Neural Language ModelsWhat followed the lecture: the measured version of the same instinct, with error bars.

How firmly this can be said

Certain What Kaplan calculated, and when
It is on video at the timestamp given. The numbers above are his.
Interpretation That the calculation is a Fermi estimate
It fits the common meaning of the term, but the label is this site’s. YouTube’s captions for all three of his lectures contain no mention of Fermi.
Not claimed That Kaplan learned the method from Fermi’s work, or had Fermi in mind
Order-of-magnitude estimation is taught to every physicist. The line marks a shared technique, the weakest kind of connection on the map, and is labelled that way.

On the map this line is marked shared technique: the later work uses a method the earlier person is known for. No contact and no citation is claimed; the pairing is this site’s.

See this line in the diagram →The best-evidenced line: Kaplan and Roberts →

Sources read on November 15, 2026. Corrections are welcome; the claim will be changed to fit the evidence.