nucleation

LIT-377: On Context-Content Uncertainty Principle

2 min read

Properties3
StatusRejected — by LIT-374 — Structural Decoupling: A Scaffold-Flow Theory of Generalization and Alignment; read in full 2026-10-01 (NOTE-325); not worth a reader's time as a source of results. Its foundation is the entropy chain rule plus an assumption, H(content) ≪ H(context), and from these it reads off a direction of inference. Of its eight theorems, one is proved only in sketch, one equivalence proposition is marked "proof skipped" and argued only in words (App. H), and the proofs of Theorems 6–8 swap the roles of content and context relative to the statements. The proof of Theorem 1 states the inequality backwards. Theorem 3 rests on D_KL(q‖p) ≥ H(q) − H(p), which is false in general. The abstract's "computational simulations" are not in the paper. The author's own v2 (June 2026) replaced the whole framework under a new title.
Tagscognition, information-theory, probabilistic-modeling
Cited as “paper” byNOTE-325 — On Context-Content Uncertainty Principle

Graph View

Li (2025), arXiv preprint, v1; replaced in June 2026 by a v2 under a different title — arXiv:2506.20699

The arXiv number now serves the v2, which this record holds as LIT-374: a different paper, Structural Decoupling, sharing no section, definition or theorem with this one. It is named here as superseded_by because the author replaced this paper with it at the same identifier, so the identifier names LIT-374. This note stays for the v1 the owner read, and stays Rejected on its own proofs.

Standing in the record

Filed on 2026-10-01 from the owner’s reading feed, where it was read on several days under its v1 title. The arXiv id now resolves to v2, Structural Decoupling: A Scaffold-Flow Theory of Generalization and Alignment. Same author, but Structural Learning Theory replaces the content entirely: width, a phase transition at width, scaffold–flow decoupling and AI safety. v2 makes claims about machine-learning systems, so it is filed in the Anthology of the SOTA as ANTH-LIT-769 (see anthology issue #180), under its v2 title. This note holds v1, the brain-theory paper. The owner directed that this paper be filed in both records, which is an exception to the rule that nothing is filed in both.

NOTE-325 is the close reading, and it placed the work: Rejected. The framework rests on the entropy chain rule plus the assumption H(Φ) ≪ H(Ψ), and each “principle” is that assumption restated. The proofs that go beyond restatement have errors:

  • Theorem 1’s proof states the inequality backwards.
  • Theorem 3’s proof uses a false KL inequality.
  • The proofs of Theorems 6–8 are about the other variable from the one in the statements.
  • One proposition is argued only in words.

The abstract’s simulations are absent. Read it, if at all, as a map of which brain theories the author thinks share a structure, not for any result. On information-theoretic accounts of inference in this record, the information bottleneck (LIT-338) and the conditional entropy bottleneck (LIT-226) make the compression claims that v1 gestures at, and they state those claims precisely.

Backlinks