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

Jared Kaplan and Daniel Roberts: a measurement, a footnote, then a model

cited in the bookRests on: in the textJared Kaplan · Daniel Roberts · 2021

The claim

In 2020 Jared Kaplan and Sam McCandlish, two former holographers at OpenAI, measured how language models improve with scale. In 2021 a footnote in Daniel Roberts and Sho Yaida’s theory book cited that paper and guessed at an explanation. In 2022 Roberts, with Alexander Maloney and James Sully, published a solvable model of those laws that used different mathematics from the guess.

Why it matters

Kaplan and Roberts have never shared a byline, an employer or, on any record found here, a room. Among the fifteen people on this map, the Anthropic founders and the Complexity Equals Action authors are joined by no documented collaboration at all. This is where they meet: in the literature, with theory answering a measurement, across competing labs. Three separate biographies would not show it.

Where to check it

  1. arXiv:2001.08361, Scaling Laws for Neural Language Models (January 2020)The measurement. The first page marks Kaplan and McCandlish as equal contributors and says they led the research.
  2. arXiv:2106.10165v2, The Principles of Deep Learning Theory, Epilogue ε, p. 391, footnotes 1 and 2The footnote. Footnote 1 cites the paper, reference [72], as an empirical study of scaling laws in transformer language models. Footnote 2 says the physics of universality “perhaps suggests” that representation group flow “may be able to explain the neural scaling laws of [72].”
  3. arXiv:2210.16859, A Solvable Model of Neural Scaling Laws (October 2022)The model. It cites the scaling-laws paper as reference [18] and the book as [16], and solves a random-feature model with random matrix theory. Sully’s affiliation line includes Anthropic.
  4. arXiv:1512.01548, Tensor Networks from Kinematic Space (2015)The second thread. James Sully was McCandlish’s coauthor in holography before he was Roberts’s coauthor on the theory of McCandlish’s measurement.

How firmly this can be said

Certain The citation, the footnote’s wording and the bylines
Each is on the page given and can be checked in a few minutes.
Stated by the authors That representation group flow explains the scaling laws
As a guess, hedged twice, in a footnote. The 2022 paper solves a different model and leaves the book’s tools to its outlook.
Interpretation That this is theory answering experiment across two labs
It is how this site reads the sequence. None of the authors is quoted describing it that way.
Not claimed Contact, collaboration or influence between Kaplan and Roberts in person
No source for it was found, and the line does not assert it.

On the map this line is marked in the text: one person’s published words cite, name or answer the other’s. The citation can be checked; how much it mattered is a reading.

See this line in the diagram →The line people ask about: Fermi and Kaplan →

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