CONTEXT JAMMING

Field notes from inside the context window.

Context Jamming / Explainers

Interactive arXiv Explainers

Famous papers from AI and physics, rebuilt as interactive research instruments. Manipulate a core diagram from each paper, then open the full explainer for the argument, equations, and source notes.

  • N° 01 · 2020arXiv:2001.08361
    Interactive sampleSelect bottleneck
    Power-law scaling frontierThree descending curves represent parameter, data, and compute limits. The selected bottleneck is emphasized and moves the frontier point.POWER-LAW FITLOG RESOURCE →LOSS →
    ACTIVE BOTTLENECK · CLOSS ≈ 1.88PPL ≈ 6.6 · FIT-DERIVED

    Empirical Laws of AI

    Scaling Laws for Neural Language Models

    How model size, data, and compute collapsed onto a small set of power laws—and reorganized the logic of training language models.

    Open full explainer
  • N° 02 · 2025arXiv:2412.14014
    Interactive sampleChange dimension
    Dimension-dependent gravitational phaseThe hand rotates among one, i, minus one, and minus i for a bare sphere. Adding the observer fixes the result at minus i.1i−1−i−1D = 4
    GRAVITYi^(6) = −1PARTICLE(−i)^(3) = iPRODUCTobserver not included

    Euclidean Gravity

    Real observers solving some imaginary problems

    Why de Sitter’s state count rotates through i—and why the observer must enter the calculation before we decide what is being counted.

    Open full explainer
  • N° 03 · 2015arXiv:1512.01548
    Interactive sampleDrag endpoints
    Boundary interval and bulk geodesicTwo draggable boundary endpoints define an interval and a curved bulk geodesic inside the disk.BULK GEODESICBOUNDARY INTERVALKinematic coordinate planeA single point at theta and alpha represents the complete interval and geodesic, not a location in the disk.POSITION θSCALE αKINEMATIC POINT
    u = 0.45 · v = 2.00θ = 1.22 · α = 0.78ONE POINT = ONE COMPLETE GEODESIC

    Kinematic Space

    Tensor Networks from Kinematic Space

    How a tensor network became a map of every possible way to probe a universe.

    Open full explainer
  • N° 04 · 2026arXiv:2601.04603
    Interactive sampleSet risk, then run
    Adaptive classifier pipelineA harmless synthetic risk score passes through the protected model and activation probe. Higher scores recruit an exchange classifier before an allow or refuse decision.EXCHANGERISK 34PROTECTEDMODELACTIVATIONPROBEPASSEXCHANGECLASSIFIERALLOW0.55 PROBE + 0.45 CLASSIFIER
    PRODUCTION CONTEXT · 94.5% PASS / 5.5% ESCALATEREADYADAPTIVE COMPUTE · SYNTHETIC ROUTE

    Constitutional Classifiers++

    Constitutional Classifiers++: Efficient Production-Grade Defenses against Universal Jailbreaks

    How Anthropic made a stronger jailbreak defense approximately 40× cheaper by moving classification into the generation stream.

    Open full explainer

Synthesis layer

Where the papers touch

Select two to four papers. The matrix distinguishes structural affinity from documented influence, revealing shared mechanisms, conceptual bridges, and the points where the comparison breaks down.

Method note. Affinity scores are editorial synthesis aids, not statistical measurements. Direct influence is shown only when supported by a citation or explicit source evidence.

Select papers for comparison

Choose 2–4 papers. When two remain, both stay selected so the comparison remains valid.

Affinity score
  1. 0–24 · Distant
  2. 25–49 · Adjacent
  3. 50–74 · Strong affinity
  4. 75–100 · Very strong affinity
Pairwise affinity among selected papers, sorted chronologically
PaperKinematic Space2015Scaling Laws2020Real Observers2024 / 2025Classifiers++2026
Kinematic Space2015Self

Recast the network as a geometry of probes, not bulk space.

Scaling Laws2020Self

Compress training behavior into empirical power laws.

Real Observers2024 / 2025Self

Insert the observer before interpreting the state count.

Classifiers++2026Self

Move full-context judgment into an adaptive compute cascade.

Scaling Laws × Classifiers++

Scaling economics becomes adaptive compute

86Very strong affinity

The scaling paper turns compute into a measurable allocation problem; Constitutional Classifiers++ applies a related economy locally, escalating only suspicious exchanges to more expensive judgment.

  • scale
  • compute allocation
  • model hierarchy
  • efficiency frontier
  • adaptive computation
  • cost-performance tradeoff
  • production optimization

Methodological rhyme

Both treat performance as an optimization problem under a finite compute budget and seek the allocation that avoids paying the highest cost everywhere.

Relationship status

Documented influence

Documented influence is narrow but explicit: Constitutional Classifiers++ cites Kaplan et al. (2020) in §5.1 for the approximation used to estimate transformer FLOPs. Jared Kaplan is also an author on both papers. This does not establish that scaling laws caused the classifier cascade design.

Productive tension

Scaling laws describe aggregate training behavior; the classifier system makes per-exchange, runtime routing decisions.

Where the analogy breaks

A classifier threshold is not a scaling exponent, and a cascade's routing policy is not a global training law.

Scaling Laws Classifiers++

Documented, limited direction: Constitutional Classifiers++ uses the scaling-law paper's compute approximation in §5.1 while estimating the marginal cost of linear probes.

Classifiers++ Scaling Laws

Retrospective comparison only: the 2026 classifier system cannot have historically influenced the 2020 scaling paper; it is a later engineering example of selective compute allocation.

Evidence notes

  • Constitutional Classifiers++ §5.1 cites Kaplan et al. (2020) for a transformer FLOP approximation.
  • The arXiv records show Jared Kaplan as an author on both papers.
  • The broader bridge from global scaling economics to adaptive runtime compute is editorial synthesis.

Selected-set synthesis

What emerges across this set

Strongest pairScaling Laws × Classifiers++

86 / 100 affinity

Bridge paperClassifiers++

77.0 average affinity

Shared vocabularyadaptive computation · measurement · compression · scale · asymptotic behavior

Most recurrent curated tags in the selected pair records

Chronology2015 Kinematic Space → 2020 Scaling Laws → 2024 Real Observers → 2026 Classifiers++

Dates order the comparison; they do not establish influence.

Central patternAcross this set, measurement is not passive: the chosen probe changes what the system can reveal and how computation is allocated.

Generated deterministically from the selected set’s curated tags.

Open method

Turn your own arXiv paper into an interactive explainer

Download the reusable Context Jamming skill that converts an arXiv paper into a complete, copy-paste Codex implementation prompt. It extracts the paper’s central thesis, maps claims to source sections, selects defensible interactive diagrams, and preserves the distinction between established results, author interpretation, and cross-domain analogy.

Built from the same editorial and technical grammar used for the Kinematic Space, Scaling Laws, Real Observers, and Constitutional Classifiers++ explainers.

  1. Source anchored

    Maps claims, equations, figures, and interactions back to the paper.

  2. Thesis first

    Finds the conceptual reversal instead of reproducing the paper section by section.

  3. Interactive by design

    Specifies deterministic diagrams that teach mechanisms rather than decorate the page.

  4. Epistemically bounded

    Separates results, author interpretation, and Context Jamming extensions.

Download the skill folder · ZIPStatic archiveDownload raw SKILL.mdUTF-8 Markdown

How to invoke it

Use the attached arXiv paper and the arxiv-to-contextjamming-explainer skill to generate a complete Codex prompt for a Context Jamming interactive explainer modeled on /kinematic.
  1. 01

    Download

    Download the skill folder and make SKILL.md available to your Codex or agent workflow.

  2. 02

    Attach a paper

    Provide an arXiv PDF, manuscript, paper URL, or sufficiently complete research notes.

  3. 03

    Build

    Run the skill, review the generated implementation prompt, and paste it into Codex with the Context Jamming repository available.

The skill produces the implementation prompt. It does not replace scientific source review or automatically prove cross-domain analogies.

What the skill requires
  • An arXiv PDF, manuscript, or paper URL
  • An optional route slug
  • An optional target-domain extension
  • Access to the Context Jamming repository or scaffold