AI Research

Calyx was created by Chris Royse. It is an association-native database and research system whose vocabulary gives this project a disciplined way to describe experimental control, selection, measurement and reversible promotion.

Our work uses Calyx as a conceptual control plane: typed lenses define measured views; association graphs and constellations connect related experiments; Lodestar selects candidates; Assay measures them; Ward gates accept or reject them; Ledger preserves provenance; and Anneal keeps promotion reversible.

CalyxGLM Lab does not copy or redistribute Calyx source. Calyx remains Chris Royse's separately licensed work, and this project gratefully credits both its authorship and the architectural ideas that inform our methodology. The reviewed Calyx snapshot carries the Business Source License 1.1; readers should consult the upstream repository for its current terms.

Sources: [2] [3]

CalyxGLM Lab is an independent, measurement-first attempt to make GLM-5.2 usable on consumer hardware while preserving model behavior. It is not an official Z.ai or Colibri project, it does not redistribute model weights, and it does not claim quality equivalence.

The published source checkpoint is described as a 753.33-billion-parameter BF16 model occupying 1.506 TB. The proposed main-plus-MTP representation is estimated at 292.03 GB, with E8 expert storage at 3.0625 bits per weight. Across six sampled real expert matrices, the reported mean synthetic-activation relative error is 0.1930, the maximum is 0.1936 and mean row cosine is about 0.981. These are component measurements, not proof of preserved language-model behavior.

Consumer-memory execution has been demonstrated: Phase 1 generated 'Hello' with 2.480 GiB peak process-group memory and zero swap. The accepted Phase 3C baseline completed an exact decode in 124.23 seconds, a 48.77 percent reduction from the first real decode baseline, while retaining the recorded safety conditions. At the reviewed public commit, the next decisive milestone is paired teacher-and-student full-logit comparison across the complete model.

Sources: [1]

Calyx RDC is an early behavior-first rate-distortion compiler design for progressive mixed-precision model representations. Its current local Rust contract core defines a representation-independent codec interface, a scalar E8/IQ3 oracle, transform scheduling, bounded-buffer ownership, routing-preserving expert queues and a validated I/O-and-compute dependency graph.

The reviewed implementation task reports twelve self-contained tests and an independent fixture passing. That verifies the present contracts; it does not make RDC a complete compiler or demonstrate retained model quality.

RDC does not yet include a model frontend, rate-distortion curve fitter, global bit allocator, progressive container writer, Metal backend or end-to-end quality evidence. Its source is not yet linked publicly; a repository link should be added only after a separate provenance and release review.

A bounded mathematical exploration asked whether a defensible geometry of numbers can be derived instead of fitted. Every positive integer has a unique vector of prime exponents. In that lattice, multiplication is vector addition, primes are coordinate directions and ordinary order is recovered by the linear functional log n = sum over primes p of v_p(n) log p.

The key rigidity fact is classical: a real-valued arithmetic function that is completely additive under multiplication and monotone in ordinary numerical order must equal c log n. The completely additive case has a short proof by interleaving powers of any two integers and applying the squeeze theorem. Erdős proved a stronger monotone-additive result in 1946.

This gives irrational log ratios a precise geometric role. For example, log 3 divided by log 2 is the limiting slope that orders the lattice directions generated by powers of 2 and 3. It is not evidence that the ordinary real line has hidden Riemannian curvature. It is an invariant conversion factor between factorization and order.

Sources: [4]

The finite experiment gave every prime p at most N an initially independent weight, fixed the weight of 2 to log 2 and imposed only that the resulting completely additive energy did not decrease between adjacent integers through N. Linear programming then bounded each small prime weight.

At N = 100,000, the allowed value of the weight of 3 was 1.0986091226 through 1.0986154946. The interval has relative width 0.0005800 percent and contains log 3 = 1.0986122887. At N = 3,000, the relative widths for primes 3, 5, 7, 11 and 13 were all between 0.0296 and 0.0346 percent. Independent Mac and Windows runs agreed to floating-point precision; the smallest constraint residuals were on the order of 10^-13.

The proposed Finite Logarithmic Rigidity Conjecture states that, for each fixed prime, the allowed interval width is N^(-1+o(1)). Eight modest cutoffs do not determine an asymptotic rate. The result is a reproducible numerical observation and a precise conjecture, not a proof or a claim of novelty after an exhaustive literature review.

Sources: [4]

The Multiplicative Spectral Sector Hypothesis asks whether nature or an engineered system contains stable states whose independently established fusion rule is multiplication, whose elementary modes factor uniquely and whose conserved energy is additive and ordered by the fusion labels. If those conditions hold, the rigidity theorem predicts one spectrum after a single scale is fixed: E_n = epsilon log n.

The elementary modes would be indexed by primes and the noninteracting thermal partition function would be zeta(beta epsilon). Prime occupation numbers would be independent geometric variables, and the partition function would diverge at beta epsilon = 1. These conclusions are conditional predictions, not evidence that such a sector exists.

The proposal is rejected if fusion is not multiplicative, factorization is not unique, binding energy breaks extensivity, measured ratios disagree with logarithmic ratios, or labels are assigned after seeing the spectrum. These safeguards prevent arbitrary relabeling from turning the idea into numerology. Julia's primon gas, the Bost–Connes system and Connes' spectral work are established precedents for exact number-theory and statistical-mechanics bridges. Recent work by Hartnoll and Yang also connects semiclassical BKL gravitational dynamics, automorphic L-functions and a dual gas of prime-labeled oscillators.

Sources: [5] [6] [7] [8]

The theorem is not new, the geometric interpretation is a framing and the finite-rate conjecture is unproved. The local scripts and full research note remain private until a separate deliberate code-publication path is approved. This proposed page reports the algorithm, parameters, platforms, bounds and solver residuals but is not yet a standalone public software artifact.

Useful next steps are exact rational dual certificates for finite bounds, sparse computations at larger cutoffs, stability theorems with noisy order constraints, generalization from primes to prime ideals and a deliberately engineered oscillator simulator with frequencies proportional to log p. None of these steps would by itself prove the Riemann hypothesis, factor large integers efficiently or establish new fundamental physics.

This page publishes reviewed summaries and reproducible public artifacts, not raw AI conversations. Conversation histories can contain exploratory claims, private machine details and superseded conclusions, so they are used as research context and reduced to owner-reviewed statements before publication.

No model endpoint, training interface or visitor-write path is exposed through the website. The AI work is adjacent systems and mathematical research and must not be interpreted as an RF observation, a radio anomaly claim, evidence that compressed-model intelligence has been preserved or evidence that arithmetic has been physically measured as the substrate of the universe.