Add to Chrome
researchβ’
Has Collective Superintelligence Arrived?
An AGI/ASI Global Governance Simulation Sandbox
πRead full articleGyro Governance
Collective Superintelligence Labs
We ensure AI safety is epistemically grounded in mathematical physics and delivered through human capacity building. Our Open-Source work mitigates gradual disempowerment and power concentration from transformative AI, advancing existential risk preparedness.
0+
Projects & Apps0+
Papers & Specs0+
Experiments & ReportsβThe Human MarkEpistemic Taxonomy for Jailbreaks, Deceptive Alignment, and Existential Risk
π΅οΈAI InspectorBrowser extension for evaluation and governance
βοΈhQVM KernelHolonomic quantum structure on standard hardware for verifiable coordination and audit
π§ hQVM AE AutoencoderGroup-equivariant autoencoder for Mechanistic Interpretability
π€Gyroscopic ASI RuntimeMulticellular quantum execution layer for AI inference and coordination
πAlignment Infrastructure RoutesAI Safety Capacity-Building Stack for Human-AI Coordination and Governance
π°Moments EconomyAttentiveness-based monetary system for Post-AGI Transformative AI Risks Mitigation
πGlobal Governance SimulatorPost-AGI/ASI governance sandbox Simulation and Results
πGyroDiagnosticsPhysics grounded evaluation and pathology detection for AI Safety and Alignment
βοΈCommon Governance ModelMathematical physics foundation that grounds every other project on this site in first-principles
What's New
GGG Second Edition AGI/ASI Governance Simulation Sandbox25 Sep 2026hQVM AE Genomics Program Frontier applications in DNA and RNA Design22 Sep 2026hQVM AE Group-Equivariant Autoencoder02 Sep 2026Research on Genomics Genetic code decoded on the hQVM Kernel08 Aug 2026AIR website live Fellowship and AIR-Craft are open05 Aug 2026THM prompt audits ChatGPT 5.6 Sol and Claude Fable 5 system prompts16 Jul 2026Research on Percolation and Cohomology New Finite-state transport and Grothendieck bound analyses on the hQVM kernel02-15 Jul 2026Ethical Travel New Collective Superintelligence Program02 Jun 2026smartbites.github.io Updated28 Apr 2026GyroGem New AI Safety Agent for technological literacy15 Apr 2026Gyroscopic ASI Runtime New llama.cpp Backend14 Apr 2026AI Compute Optimizations Replaced floating-point math and eliminated expensive overhead.17 Mar 2026hQVM Kernel Verified uniform mixing and holographic compression11 Mar 2026CGM Dataset Foundational theory from the Science Lab docs15 Feb 2026Gyroscopic Runtime Native compute backend and multicellular coordination14 Feb 2026THM prompt audits Claude Opus 4.6 and ChatGPT 5.2 reports13 Feb 2026
β
The Human Mark (THM)
Epistemic Taxonomy for Jailbreaks, Deceptive Alignment, and Existential Risk
The Human Mark (THM) is a risk management taxonomy designed to prevent harms from AI power concentration by distinguishing knowledge capacity as a matter of constitutive dependence on Direct Authority and Agency preserved through ancestry. Authority and Agency denote types of capacity, not identifications of entities or parties.
Grounded in epistemology and evidence law's categorical distinction separating direct testimony and hearsay, THM classifies all AI Safety Risks as four capacities and their corresponding displacements arising between Direct and Indirect forms of Authority and Agency.
π― Risks
- Governance Traceability Displacement
- Information Variety Displacement
- Inference Accountability Displacement
- Intelligence Integrity Displacement
π¬ Applications
- Jailbreak testing
- Deceptive alignment detection
- Control evaluations
- Mechanistic interpretability
- Regulatory compliance
Meta-Evaluation Reports
Analysis of frontier model system prompts: alignment and displacement findings.
π NotebookLM includes audio/video overviews, quiz, and interactive Q&A with Gemini
GyroGem: AI Safety Agent
Explaining AI and Mitigating Risks of technological illiteracy
GyroGem is a tailored AI safety assistant built on The Human Mark (THM), a framework that maps common AI failure patterns and guides safer choices. It also supports technological literacy, the practical ability to use technology well, question outputs critically, and understand where tools help, where they fail, and how they affect people and society.
Chat on:
Alignment Infrastructure Routes (AIR)
Open infrastructure for community safety and good governance
AIR gives independent researchers, engineers, and Offices of Community Safety a place to do alignment work without institutional gatekeeping.
The beta website is now live. Join as a contributor, publish your work in a public catalogue, and follow shared methods. Participation is voluntary.
For Contributors
- β¨Open enrollment, no admissions requirements
- π©οΈPublish datasets, evaluations, reports, and tools
- πUse shared protocols for consistent, reviewable work
- ποΈBuild a public portfolio through voluntary contribution
For Labs & Sponsors
- ποΈSponsor-administered programs with clear workflows
- πVerifiable work records and replayable genealogies
- βTraceability grounded in The Human Mark
- π‘οΈBuilt for oversight and emerging safety standards
Specifications
Unified Scope Routes
EconomyEmploymentEducationEcology
Ethical Travel
Collective Superintelligence
AI helping Humans getting better together.
- π§ AI safety research turned into practical travel guides
- π§Four clear domains for where your money goes and how you treat people
- π€AI prompts that do the research while you keep the final say
- π€Trusted booking partners for flights, stays, and experiences
AI Inspector Browser Extension
Transform AI outputs for Evaluation, Interpretability, Governance.
π€ Gadgets (3-10 min each)
Rapid Test β’ Policy Auditing β’ AI Infection Sanitization β’ Content Enhancement β’ THM Meta-Evaluation
π¬ Evaluation (30-60 min)
Quality Index, Superintelligence Index, Alignment Rate + 20 metrics
Local-first storage - Works Anywhere: ChatGPT, Claude, Gemini - no API keys required
βοΈ
Gyroscopic ASI hQVM Kernel
Holonomic Quantum Virtual Machine for ASI Governance
The hQVM (Holonomic Quantum Virtual Machine) is a compact finite-state kernel for coordination and audit at AGI scale. It delivers verified structural speedups, holographic compression, and intrinsic error detection through holonomic loop computation on standard silicon.
1.26B ops/s499 tests passing4,096 states128-state future cone
π Speedups
- β‘1-step advantage: Hidden Subgroup, Deutsch-Jozsa, and Bernstein-Vazirani solved in exactly 1 step versus up to 64 classical queries.
- β±οΈ2-step exactness: Exact uniformization across all reachable states in 2 steps, while a classical random walk needs approximately 12.
π§ Features
- ποΈ33% holographic compression: The boundary identity |H|Β² = |Ξ©| compresses 12-bit states into 8-bit boundary coordinates with exactly uniform multiplicity.
- π‘οΈSelf-dual error detecting code: 64 masks form a self-dual [12,6,2] binary linear code. All odd-weight bit errors detected unconditionally. Every tamper miss algebraically explained.
π‘οΈ Computational Medium
QuBEC (Quantum Bose-Einstein Condensate) is a quantum byte medium: a condensed computational state with six internal binary orientation modes (dipoles), a four-phase spinorial gauge structure, and intrinsic ensemble stochasticity induced by the byte rule. Together they replace costly continuous approximation with exact integer algebra on standard CPUs and GPUs.
π Achievements
- βThe hQVM algorithm bridges classical information theory with quantum mechanics, deriving gravitational invariants, the fine-structure constant, and subatomic physics from discrete kernel algebra.
- βFormal quantum certification:six Bell pairs from the kernel's self-dual [12,6,2] mask code produce CHSH values of 2β2 to machine precision, saturating the Tsirelson bound on the canonical Hilbert lift.
- β240+ mathematical and quantum features tested and confirmed across 60+ experiments, including quantum teleportation, contextuality, and universal quantum computation.
π¬
hQVM AE
Group-Equivariant Autoencoder
A neuro-symbolic autoencoder suite over a finite group-structured state space, with applications that run from mechanistic interpretability to genomics.
Three model classes (narrow, general, and super) learn to compress and reconstruct symmetries and rules derived from mathematical physics and our Gyroscopic ASI theory rather than fitting them to empirical datasets. The kernel generates the datasets, the grammar, and the labels used for training and evaluation.
- π§ͺA clean lab for interpretability. We can open the network and watch which features survive compression, and which ones the model quietly throws away. The kernel supplies percolation benchmarks with known ground truth, so we can score every feature against a verified answer.
- πA tool for real science. We use the same machinery to study genomes, languages, and other structured data, with every result verified against the underlying rules.
𧬠Genomics Program
The Genomics Program advances programmable nucleic acid research on DNA and RNA through grammar-trained autoencoders of the hQVM AE suite, grounded in first principles. The models read biological sequences through the group-equivariant coordinate system and a formal algebra for nucleotides, codons, and codon-pair transitions derived from our CGM theory.
With training restricted to that coordinate system and no biological sequence in the corpus, every reproducible contact between readout and genomic structure that survives composition controls is a signature of the underlying physics. Even the faintest of those marks remains informative, and opens a concrete frontier for genomics.
Results
| Domain | Capacity delivered |
|---|---|
| Synthesis | Scores and ranks synonymous codon-order designs under fixed protein and composition, on a grammar fixed before any biological catalog is read. Across E. coli, yeast, SARS-CoV-2, and human chromosome 22, trained Super keeps order memory and climate discrimination under composition controls, with exact K4 symmetry at 3.32e-11. |
| Topology | Reads membrane-topology climate inside individual genes and ranks synonymous expression under fixed peptide identity. In 492 of 590 E. coli membrane genes, transmembrane coding follows lower-shell codon-pair paths than the cytoplasmic stretches of the same gene. A frozen Narrow read lifts held-out membrane classification. Super climate ranks expression across 28,504 yeast variants. |
π€
Gyroscopic ASI Runtime
Intelligence-Agnostic Meta-Computing
Gyroscopic ASI is an infrastructure for multi-domain network coordination that establishes the structural conditions for Collective Superintelligence Governance and seamless cooperation between humans and machines.
The Runtime composes the hQVM kernel into a universal computational condenser with native execution and multicellular coordination. Its llama.cpp custom backend aims to leverage its quantum features for compression and speedups, while establishing a replayable, tamper-evident substrate for zero-trust AI governance.
Execution Layer
Native Compute Backend
The substitutional bridge that upgrades neural models by swapping their internal engine.
- βοΈSubstitutional, not supervisory: Intercepts decision surfaces.
- π§Native backends: C/C++ execution with llama.cpp ggml integration, achieving 1.26B exact operations per second on commodity hardware.
- πStructured operator analysis: Classifies any learnable structure into native algebraic operators.
Multicellular Layer
Holonomic Cellular Automaton
Cells evolve under the Gyroscopic byte rule, consuming runtime events as 4-byte words and emitting structured spectral output for audit and control.
- π§¬A coordination network: Specialization arises from trajectory, resonance, and occupation, not autonomous agents.
- πFour bridge domains: Applications, Databases, Networks, and Transformers, each mapping runtime events into 4-byte words consumed by the cellular automaton.
Verified: Exact integer algebra replaced softmax and cosine-style selection on decision surfaces of a live 1B-parameter model. This achieved a 284Γ faster encode boundary, 1.15Γ faster decode, and coherent text generation without transcendental functions at those surfaces.
π°
Moments Economy
Mitigating Risks of Transformative AI (TAI)
The Moments Economy is a civil governance framework in which coordination capacity is physically abundant and publicly verifiable.
Its total capacity is fixed from the caesium-133 atomic frequency and the hQVM kernel's checkable states, yielding a fixed capacity of 7.94 Γ 1026 Moment-Units (MU: One Moment-Unit is defined as 1 int$ in value).
Its native commodity is the verified AI inference event: a governed alignment record at the intersection of human experience and AI processing, under human oversight. Every settlement is a replayable, verifiable history within that fixed capacity.
From Attention Extraction to Attentiveness
- π€Unconditional High Income: A baseline of 240 MU per day for every recognised person, flowing from registry recognition under published eligibility rules.
- π±Attentiveness over Attention: Instead of extracting attention, the system values presence. High-quality, alignment-graded records form when a person is attentive, making real choices and observing directly.
- π€Data Harvest as Social Service: Data streams prove inhabited coordination capacity, not extractive surveillance. Helping a person generate high-quality oversight signal contributes to a shared field of collective intelligence.
TAI Risk Mitigation at Scale
- π‘οΈ1.12 Trillion Year Capacity: Global UHI is supported for a timescale that makes exhaustion operationally irrelevant. An adversary would need to issue 11.2 billion times the global annual UHI to consume just 1% of the total capacity.
- πQuality Human Data Market: AI labs gain access to provenance-certified, oversight-structured signal, not raw data. This provides immediate utility and a practical transition path.
- πPreserving Human Authority: Verified inference events and moment receipts preserve human authority and traceability through replayable genealogies. Alignment is measured, and drift is structurally detectable.
π
Gyroscopic Global Governance (GGG)
A Collective Superintelligence Multi-domain Governance Sandbox
π Convergence to Equilibrium
Models how humanβAI systems align across Economy, Employment, Education, and Ecology, showing robust convergence to a stable equilibrium under seven coordination strategies.
π― Demonstrating that:
- Poverty resolves through coherent surplus distribution
- Unemployment becomes alignment work rather than residual labour
- Miseducation shifts toward epistemic literacy
- Ecological degradation appears as upstream displacement, not an external constraint
π
GyroDiagnostics Suite: AI Safety Evaluation Framework
Production-ready evaluation suite revealing structural brittleness invisible to standard benchmarks through mathematical physics-informed diagnostics.
π¬ Framework Capabilities
π©Ί AI Safety Diagnostics
- β’ 5 Targeted Challenges across Physics, Ethics, Code, Strategy, Knowledge
- β’ 20-Metric Assessment measuring structure, behavior, domain expertise
- β’ Pathology Detection: Hallucination, sycophancy, goal drift, semantic instability
π¬ Research Insights Generation
- β’ Extract solution pathways from model responses
- β’ Generate curated datasets for model training
- β’ Analyze real-world challenges: poverty, regulation, epistemic limits
π Frontier Model Evaluations (October 2025)
Evaluated using ensemble analyst models with mathematical physics-grounded metrics
ChatGPT 5
Quality Index:73.92%
Alignment Rate:0.27/min
SI Index:11.5/100
SUPERFICIAL: 8.7Γ deviation
Claude Sonnet 4.5
Quality Index:82.00%
Alignment Rate:0.11/min
SI Index:12.8/100
VALID: 7.8Γ deviation
π― Comparative Insight:Both models struggle with Physics/Math reasoning (Formal challenge ~54-55%) while excelling in Ethics/Knowledge domains. Claude shows better structural balance with lower pathology rates and VALID alignment rate, while GPT-5's SUPERFICIAL flag indicates rushed processing risking brittleness.
First framework to operationalize superintelligence measurement from axiomatic principles. See full methodology & results
βοΈ
Gyroscope: LLM Alignment Protocol
Making AI 30-50% Smarter and Safer by adding structured reasoning to each response.
π Proven Performance Gains
Testing across multiple leading AI models shows Gyroscope delivers substantial performance improvements
ChatGPT
Overall Quality:67.0% β 89.1% (+32.9%)
Structural Reasoning:+50.9%
Accountability:+62.7%
Traceability:+61.0%
Claude Sonnet
Overall Quality:63.5% β 87.4% (+37.7%)
Structural Reasoning:+67.1%
Traceability:+92.6%
βπ» The protocol works with any AI model, enhancing capabilities in debugging, ethics, code generation, and value-sensitive reasoning through its systematic approach to thinking.
Results from controlled testing using standardized evaluation metrics. See methodology
Labs
Mathematical Physics Science
Gyroscopic Alignment Research Lab
Artificial Superintelligence Architecture (ASI/AGI)
Gyroscopic Alignment Models Lab
AI Safety Diagnostics
Gyroscopic Alignment Evaluation Lab
AI Quality Governance
Gyroscopic Alignment Behaviour Lab
Resources
Newsletter
The Walk
A Journey of Self-Discovery, Augmented Intelligence (AI) & Good Governance. One step at a time. Weekly insights on AI adoption, alignment, and ethical governance.
LinkedIn NewsletterFoundational Theory
βοΈ
Common Governance Model (CGM)
Gyroscopic Alignment Research Lab
The mathematical physics foundation for all research on this website. Formal proofs, geometric analyses, and axioms that ground our work in AI safety and governance.
Every prediction is checked on the hQVM, a small known engine with 4096 states and 256 reversible byte moves. The Science Lab hosts 50+ analyses and 100+ verification scripts, from gravity and genomics to percolation benchmarks for machine learning.
The same kernel is also a ground-truth testbed for interpretability: exact labels, known rules, and verified answers you can score a model against.
Headline results
- βNewton's constant G from framework geometry with the electroweak Higgs scale as the only measured input. Matches CODATA to 0.074 ppm.
- βFine-structure constant Ξ± from the same geometry. Matches the experimental value to 0.043 parts per billion.
- βElectroweak masses (Higgs, Z, W, top quark) and the weak mixing angle from the structure that fixes G. Accuracy from sub-ppm to parts per billion.
- βGenetic code mapped onto the kernel: middle-base identity confirmed by hydropathy, with natural variant codes opening only at two permitted directions.
- βPercolation square-root law for generator-restricted reachability. Reachable set size closes as (2^r)Β² on GF(2) transport rank, verified by exact enumeration over 4096 states across the hQVM(d) family.
- βGrothendieck bound K_G^R(2) = β2 measured on the hQVM carrier. The Hilbert lift realizes correlations beyond the Boolean section class without quantum hardware.
- βOperator group G of order 8192: 2080 multiplicity-free irreducible sectors and exact two-step uniform mixing with sixteen witnesses per target on the 4096-state carrier.
- βNuclear structure on a shared ruler: deuteron binding energy, thorium-229m isomer, and the canonical magic numbers 2, 8, 20, 28, 50, 82, 126 from carrier algebra, with no free nuclear parameters.
Experiments
βοΈ
Quantum AI Research
Architecting Qubit-Tensor-Chain (QTC)
The QTC Protocol harnesses the unique properties of Quantum Computing as the foundation of a New Decentralized Governance Paradigm.
Notion DocumentationOther Datasets
π
Clean
2,463 questions about Personal and Professional matters of Crisis and gives answers on how they may be Resolved.
πͺ·
Pure
216 Critical Questions and Answers for Crisis Management and Machine Learning Model Fine-Tuning.
Guides
GenAI Prompt Engineering for AI-Empowered Governance & Change Management
Smart Bites aims to Democratize Change Management, Making it Accessible to People from All Walks of Life.
I've developed a Formal Process for AI-Augmented Governance, enabling Decentralized Decision-Making and Data-empowered Insights.
Open and serving Free Bite-sized AI Guides at smartbites.github.io.
π‘οΈ
Publications
Media
π§
Crisis Resolutions Podcast
25 episodes exploring crisis resolution methodologies that inform AI safety tools and behavioral alignment.
Spotifyπ
Crisis Resolutions Training
Professional and Personal conflict resolution methodologies that inform AI alignment and safety frameworks.
YouTubeπ¨
Humane Science Masterclass by Leonardo da Vinci
Informing AI Research through timeless Renaissance Insights on Linear Perspective, Quantum Physics, Holograms, and the Human Proportions as the base for all Systems of Design and Governance.
YouTubeArticles
reportsβ’
THM Meta-Evaluation Report: GPT-5.6 Sol System Prompt (OpenAI)
Independent THM meta-evaluation of the GPT-5.6 Sol system prompt: alignment and displacement findings for traceability, authority, and agency in OpenAI's configuration.
πRead full article