The situation is measured three ways. A census of ten public pre-training corpora shows that the supply of long, coherent, low-boundary-density text has collapsed to a rounding error, and that the newest corpora are the most heavily marked. A converter study shows the markup is faithful — which makes the cue more trustworthy, not less, and closes the obvious escape route: after serialization no document has zero markup, so there is no null level left to reweight toward. And a paired-serialization probe on a small base model puts the framing under direct reading-side test.

Swapping a boundary's notation while keeping its typographic form changes the model's use of long-range context by a measured zero — the operative cue is the announcement, not the sigil. Deleting the announcement itself makes the following prose measurably harder to predict, and the model recovers none of the loss from the long context that contains the evidence. The marker is informative, and the inference that should substitute for it is absent.

What follows is deliberately conservative: format augmentation rather than format replacement — present the same structure in multiple notations so the sigil stops being a shortcut — and long-context slot substitution as the first serious experiment, because the long-context stage is the one place where the training sequence is the document, and it is currently fed the most heavily marked corpus anyone has built.

• 00:18 — Summary
• 03:00 — Nothing left to infer
• 06:53 — Nobody decided this
• 10:05 — What the corpora contain
• 13:49 — The cue that cannot be reweighted
• 19:26 — The one stage where the sequence is the document
• 24:30 — A first measurement
• 32:07 — Format augmentation
• 37:07 — The program
• 42:31 — Coda: record it, then build it

The substitution happens upstream of every control the field exercises over training data. Mixture weights choose how often to sample the converted document; filters choose whether to keep it. Both take the corpus as given, and the corpus was given by an extractor. No reported quantity would change if a different extraction library had been chosen.

This proposal sets out three studies to establish whether that matters, ordered so the cheapest can end the inquiry. The first needs no models and no training: run current converters over two document classes and count the structural elements in the output with no counterpart in the source. If converters turn out to be conservative on continuous prose, the concern dissolves. The second reads public checkpoints for a predicted divergence between benchmark performance and long-form likelihood. The third — the only expensive one — is a three-arm matched pre-training run separating faithful conversion from imposed hierarchy, which no existing experiment does.

• 00:21 — Summary
• 03:40 — The gap
• 05:42 — The hypothesis
• 08:29 — Why the gap has persisted
• 10:40 — What has to be distinguished
• 12:37 — The program
• 17:51 — What would follow, and what would not
• 19:32 — Why now rather than later
• 23:08 — Relation to existing work
• 25:01 — What is already known about the model side
• 26:53 — What this would produce

Four results follow. A characterization theorem fixes the log-Poisson law with explicit parameters across all multipliers with finite lattice moments. A classification theorem locates that class inside the log-infinitely-divisible family and identifies the mechanism by which every rival sub-family fails the symmetry. A stability theorem gives sharp constants, and a propagation theorem carries the bound to the multiplier distribution at the exact rate, with a matching lower bound.

Beyond independence the picture splits, and the paper is precise about where. The classification extends exactly at the level of asymptotic statistics and provably not at the level of laws: an explicit stationary ergodic Markov multiplier satisfies the symmetry with a non-log-Poisson marginal, exchangeable multipliers collapse back to the i.i.d. case, and finite-state Markov multipliers cannot satisfy the symmetry at all.

The argument here is that they are one phenomenon: the em dash is markdown leaking into prose — the smallest surviving unit of a structural orientation acquired from markdown-saturated training data, and amplified afterwards by fine-tuning. The paper lays out that genealogy in five steps, from corpus composition through the dash's dual status as both punctuation and structural joint.

Then it tests it. Twelve models from five providers were instructed to suppress markdown. Headers, bullets, and bold vanish; em dashes persist — except in Meta's Llama models, which produce none at all. Rates run from 0.0 per thousand words to 9.1 under active suppression. A three-condition gradient shows that even explicit prohibition of the dash fails to eliminate it in some models, and a base-versus-instruct comparison finds the tendency already present before RLHF. The conclusion reframes em dash frequency as a diagnostic of a particular fine-tuning procedure rather than a stylistic defect.

• 00:18 — Abstract
• 02:27 — 1. Introduction
• 07:22 — 2. Background and Related Work
• 13:58 — 3. The Em Dash Genealogy
• 24:30 — 4. The Two Discourses
• 27:41 — 5. Empirical Evidence
• 42:04 — 6. Discussion
• 50:58 — 7. Conclusion

Between 2022 and 2025 the same requirement surfaced independently across five research programs in machine learning, mathematics, neuroscience, and psychology — not an account of what consciousness is, but of what it needs. Hinton's mortal computation and Hoel's disproof arrive at it from opposite directions; reconsolidation work in memory science had been circling it for a century. What they agree on is a system that cannot process information without being changed by it.

The essay takes the three strongest objections seriously — test-time training, whole-brain emulation, neuromorphic hardware — and says plainly what would falsify the claim. If it holds, the obstacle to machine consciousness is not scale. It is that a computer can read a value without disturbing it, and we cannot, and that failure may be the whole of what it is to be someone.

• 03:08 — Five Paths to the Same Conclusion
• 11:27 — A Century of Clues
• 14:05 — Karim Nader and the Read That Writes
• 18:23 — One Concept, Understood Immediately
• 22:12 — Three Objections Worth Taking Seriously
• 26:18 — What It Means If They’re Right
• 27:11 — What Could Prove Them Wrong
• 30:18 — The Inversion

The lily is the way in. Study one and what registers is that every part of it, at every scale, is opening — as though opening were its entire purpose. From there the essay turns to what a fractal actually is, which is not a self-repeating pattern but a shape with a non-integer dimension, and finds in the Julia set at 1.2683 dimensions the family that sits where the lily sits: in the liminal space between a line and a plane.

What follows is an argument about looking. Self-similarity is offered as the reason the subject resists comprehension and rewards attention, and the essay ends where it began, on whether the reader can now see it.

• 00:49 — The Lily
• 01:26 — It’s Between Dimensions And It’s In Between Her Legs
• 01:56 — The Julia Set
• 03:24 — How Fractal Is She?
• 04:18 — Her Lily’d Nature
• 05:26 — In Full Bloom
• 06:39 — Do You See?

Size is personal preference and, isolated from the whole, a meaningless vector for aesthetic analysis. If one factor has to carry the argument, it is not size but the hang: big, small, asymmetric, tuberous, plump, flat — these are variables that combine into the composition, and the composition is what is being responded to.

The essay argues the case through movement. A body with exceptional hang is one whose smallest change of posture unlocks a new view, and the pleasure is in the transformation rather than in any single frame — a delicate balance implying that this particular view will not be seen again. Sections on discernment and on suboptimal hang do the work of distinguishing the claim from mere preference, which is what taste requires and what the meme skips.

• 01:39 — Movement And Transformation
• 03:35 — No Less Expansive
• 05:25 — Careful Discernment
• 07:10 — Suboptimal Hang
• 07:57 — Movements That Move Us