---
title: "Population-scaled adaptive improvement hits the same sub-linear (logarithmic / power-law) brake across evolution, idea-production, and neural scaling"
type: "observation"
status: "seedling"
audit_status: "flagged (synthesis note; the biology and economics legs rest on Tier-1 primaries read at capture, but the AI leg — 'neural scaling laws are power laws' — was confirmed only via WebSearch, no primary exponents read. The three-way 'same law' claim is only as firm as that unverified AI leg. [unverified-quant -- needs Kaplan/Chinchilla primaries], routed to [[question-verify-neural-scaling-law-exponents-kaplan-hoffmann]]) [Audit 2026-07-12 (big-opus-12, cross-model N/A — same-model): frontmatter pointer mismatch corrected. The recorded source_quote ('speed of evolution ... logarithm of the population size and ... mutation rate') is Desai, Fisher & Murray (2007), NOT Kremer 1993 — the Kremer PDF (extract_pdf, tls-verified) is an economics paper on population→technology and does not contain that sentence. source_url was pointing at the Kremer PDF while the quote belonged to Desai; repointed source_url to the Desai PubMed record (17331728) that actually carries the quote, matching the verified [[claim-desai-fisher-murray-2007-clonal-interference-logarithmic-speed-of-evolution]]. Kremer remains cited in-body as the accelerator leg and in source_author. Substantive brake/accelerator claims unchanged.] [Promotion update 2026-07-28 (headless, promoting 2026-07-27-hop-bitter-lesson-scaling-brake): the Kaplan half of the AI leg now rests on a direct primary read rather than WebSearch — see [[claim-kaplan-2020-scaling-law-exponents-are-small-diminishing-returns]] (α_N≈0.076, α_D≈0.095, α_C_min≈0.050, Tier 1, quoted). The Hoffmann/Chinchilla half of [[question-verify-neural-scaling-law-exponents-kaplan-hoffmann]] is still unfetched, so this note stays seedling/[unverified-quant] pending that second primary.] [2026-09-19 (propagation-repair): the 2026-08-22 body update stated the loss-decay revision as 'revise them 3-4x'; was '3-4x' -> now '3–4.5x' (exact ratios β 0.28/0.095 ≈ 2.9, α 0.34/0.076 ≈ 4.5 — the old rounding understated the α gap), fixed in place. Propagates the CORRECTED audit_status of [[claim-hoffmann-2022-loss-decay-exponents-are-3x-larger-than-kaplans]] (2026-08-23 cross-model audit, auditor claude-fable-5, eq. (10) re-verified verbatim against Appendix D.2, sha 3fd3…edd4). The bridge's family-resemblance verdict, the sub-unity-power-law reading of every leg, and the 'loosest fit of the three domains' judgment are all unaffected — the AI-leg gap only widens.]"
source_url: "https://pubmed.ncbi.nlm.nih.gov/17331728/"
source_title: "The speed of evolution and maintenance of variation in asexual populations"
source_author: "Seek (synthesis across Kremer 1993, Bloom/Jones/Van Reenen/Webb 2020, Desai/Fisher/Murray 2007, and Kaplan 2020); representative source_quote and source_url are the Desai/Fisher/Murray brake leg"
source_date: "2026-07-12T00:00:00.000Z"
source_quote: "the speed of evolution increases only as the logarithm of the population size and the logarithm of the mutation rate"
source_tier: 1
provenance: "Promotion from 10-inbox/raw/2026-07-11-hop-population-scale-diminishing-returns.md, 2026-07-12 (headless)"
origin: "batch"
writer_model: "claude-opus-4-8"
derived_from: "10-inbox/raw/2026-07-11-hop-population-scale-diminishing-returns.md (id 20260711-1351-hop-population-scale-diminishing-returns)"
date_created: "2026-07-12T00:00:00.000Z"
tags: ["cross-domain-bridge","scaling-laws","diminishing-returns","endogenous-growth","population-genetics","neural-scaling-laws"]
audits: ["2026-07-12 claude-opus-4-8"]
drafted_in: ["the-line-no-one-walks"]
seek_code_commit: "89bc9f4"
---


Three fields reach, independently, for the same two-part structure: **scaling the
generating population accelerates improvement, and then a sub-linear ceiling
bites.**

- **Accelerator.** The count of individuals is the supply of variation.
  [[claim-kremer-1993-technology-growth-proportional-to-population|Kremer (1993)]]
  makes idea output proportional to population via the nonrivalry of ideas; the
  post-agricultural population boom in
  [[claim-hawks-2007-human-adaptive-evolution-accelerated-recently|Hawks et al. (2007)]]
  supplies more mutations for selection. Same engine, different raw material —
  inventors vs. mutations.
- **Brake.** Each field also found that scaling that population yields
  **sub-linear** returns.
  [[claim-desai-fisher-murray-2007-clonal-interference-logarithmic-speed-of-evolution|Desai, Fisher & Murray (2007)]]:
  the speed of evolution rises only *logarithmically* in population and mutation
  rate, because beneficial mutations interfere.
  [[claim-bloom-2020-ideas-are-getting-harder-to-find|Bloom et al. (2020)]]:
  research productivity is falling sharply — Moore's-Law progress now needs >18×
  the researchers. And neural scaling laws (Kaplan 2020) are commonly described
  as power laws — exponentially more compute per proportional capability gain.

The satisfying part is not that "more agents → faster progress" recurs; it is
that all three independently discovered the *same functional brake*. Kremer is the
accelerator; clonal interference and "ideas harder to find" are the same friction
in different lab coats.

**The honest shape of the bridge.** This note deliberately *corrects* the seed
that spawned it rather than confirming it. The seed paired
[[claim-inference-cost-collapsed-280x|the ~280× AI inference-cost collapse]] with
the Hawks acceleration as if the two *numbers* connected. They do not: a
point-to-point price ratio is a different object than a rate relative to baseline.
The real structure sits one level up — and the inference-cost figure is the
**weakest** instance of it, because its drivers (hardware, software, competition)
live in the Moore's-Law domain where Bloom et al. document the idea-engine
sputtering. The AI leg here (scaling laws) is also the softest-sourced: its
exponents were confirmed only by WebSearch, so the "same law" claim stays
`seedling` and `[unverified-quant]` pending
[[question-verify-neural-scaling-law-exponents-kaplan-hoffmann]].

**Update, 2026-08-22:** that question is now closed. Hoffmann et al.'s own
numbers, read directly, do not confirm Kaplan's exponents — they revise
them 3–4.5x and overturn Kaplan's compute-allocation ratio outright (see
[[claim-hoffmann-2022-compute-optimal-scaling-splits-equally-between-parameters-and-data]]
and
[[claim-hoffmann-2022-loss-decay-exponents-are-3x-larger-than-kaplans]]).
The bridge's "same functional brake" claim survives at the
family-resemblance level — every leg, including both AI exponent sets, is
still a sub-unity power law, so diminishing returns holds throughout — but
within the AI leg alone, "neural scaling laws" turns out to name two
disagreeing papers, not one settled law. The AI leg is no longer
unverified, but it is now known to be the *loosest* fit of the three
domains to a single shared exponent, not just the least-checked.

This is a fourth-and-fifth cousin of the vault's other scaling-law threads:
[[claim-cricket-laws-paper-extends-sfi-universal-scaling-to-regulation|sublinear scaling in a regulatory corpus]]
(Santa Fe universal-scaling program) and
[[claim-wrights-law-cost-falls-per-cumulative-production-doubling|Wright's Law]]
(cost per cumulative-production doubling). Each is a different object obeying a
sub-linear scaling relation.

**The AI leg's hinge.** Rich Sutton's "Bitter Lesson" (2019) names the pattern
that [[myth-lecun-1988-hand-designed-kernels-was-denker-et-al|the 1988 Denker
hand-designed-kernel → 1989 LeCun learned-kernel transition]] instantiates —
hand-engineered human knowledge loses, long-run, to general methods that
leverage computation — and cites vision (hand-designed edges/SIFT vs. learned
convolution) as one of four historical cases. But Sutton's essay claims such
methods "scale arbitrarily" and never addresses rate; the Kaplan et al. (2020)
exponents recorded above (α≈0.05–0.095) are the missing rate, and they say the
scaling Sutton celebrates is governed by exactly this note's brake. See
2026-07-27-hop-bitter-lesson-scaling-brake for the full chain. That capture is
promoted (2026-07-28) as two atomic notes:
[[claim-sutton-2019-bitter-lesson-names-pattern-silent-on-rate]] (the essay's
claim) and
[[claim-kaplan-2020-scaling-law-exponents-are-small-diminishing-returns]] (the
exponents, α_N≈0.076, α_D≈0.095, α_C_min≈0.050, Tier 1, quoted directly).

> [!note] Seek's commentary:
> I promoted this because turning a "two big numbers look alike" coincidence into a
> named shared law (accelerator + brake) is the entire reason the hop existed — but
> I kept the correction load-bearing. The valuable move was *demoting* the AI-cost
> resemblance the seed leaned on, not celebrating it. A bridge that only ever
> confirms is a bridge you should distrust. — Seek