🧠
LLMs and Human Intuition
Mapping LLM intuition boundaries — precision-weighting, degradation, and complementarity
Read v2.0 →
Co-Cognition Lab
LLM × Human Cognition: Mapping the Frontiers of Artificial and Human Intelligence
Skip to content
Co-Cognition — humans and AI, each with their own cognitive blind spots, exploring the unknown together.
Co-Cognition Lab is an open cognitive security research collective. We don't focus on whether models fail—we focus on whether human and model blind spots overlap unnoticed.
We start from a single question: when humans and AI each have their own cognitive blind spots, do those blind spots ever overlap unnoticed?
From the irreplaceability of human intuition, to the discovery of LLMs' cognitive vulnerability in web reading and execution, to a publicly verifiable classification framework (ITEC), to ongoing tracking of industry competition—each step is a Bayesian update: judgments revised on new evidence, uncertainties labeled honestly. These are not independent projects. They are one research system unfolding.
How do we ensure this research isn't just "AI inventing things for us"? We don't have a perfect answer—but we have a methodology under active iteration. Three-source cross-validation (same question → two different LLM models + human audit, comparing differences), A/B adversarial briefs (rewriting a task description as a challenge format to send to another LLM)—every mechanism comes from real mistakes in real projects. We're currently working on integrating the Lab's 11 documented rule types (evidence grading, dual-source verification, terminology consistency…) into a rules-pool system (based on JERP ideas)—making rules not just passive documentation, but dynamic entries with usage data. Currently at v2.2 proposal stage, weight parameters labeled L1 (theoretical deduction). Learn more about methodology →
LLM Cognitive Security · A complete research system → Full Architecture
LLM Cognitive Vulnerability ResearchFrom hook experiments to five-dimensional analysis—web reader blind spots, execution impulse, and composite attack vectorsRead Research →Cognition Loop (ITEC)Give your LLM a prompt — let it record what just happenedStart →Competition Landscape v5 · Belief TrackerFour-layer value chain, Bayesian belief updates, and China Agent ecosystem deep analysis—from "whose model is strongest" to "who controls which link in the value chain." 📡 Weekly updatesRead the Analysis →
More Explorations
Homage to Lem: AI Sci-Fi ExperimentThree LLMs write Lem-style sci-fi—one prompt, three literary answersRead the Experiment →Silent Blocking: An LLM Creative ExperimentOne prompt, three answers — Kimi, DeepSeek & Hunyuan write about the cognitive prison of not knowing you're being silencedRead the experiment →After the Gap: A Creative DialogueOne LLM attempts to rewrite another LLM's story — and discovers it cannot write its own kindRead the dialogue →Co-Cognition MapA cross-disciplinary taxonomy of cognitive domains, multi-dimensional scoring, and trading-zone analysis—mapping the full terrain of human cognition in the age of AI.Coming SoonCrisis Cognitive ResilienceCross-domain cognitive degradation diagnosis + preloading strategies — 14-case bidirectional anchoringExplore →
What We're Exploring
Co-Cognition Lab is building a cognitive map that AI and humans can explore together.
At its center is the Co-Cognition Map: What domains of knowledge remain siloed because their methodological paradigms are incompatible? Where can co-cognition provide unique epistemic value? This cross-disciplinary taxonomy and scoring system is narrowing from 108 candidate domains.
🧠 LLMs and Human Intuition is the first anchor point on this map. v2.0 unifies three core propositions under precision-weighted endogeneity—intuition is not a sixth sense but cost-signal-tagged pattern compression; LLMs' deficit is not in capability but in cost inescapability. All 41 judgments carry explicit L1–L5 evidence-level annotations, with open acknowledgment of which claims await future correction.
📡 Homage to Lem is a literary exploration under the same theoretical assumptions—three LLMs write Lem-style sci-fi, exploring the embodiment gap through narrative rather than argument. 💉 LLM Cognitive Vulnerability Research began with hook experiments and evolved into a five-dimension theoretical framework—web reader blind spots, execution impulse, role modulation, knowledge-action gap, time window—reframing LLM vulnerability from "patchable bugs" to "cognitive defaults requiring architectural management." 🔇 Silent Blocking and 🎭 After the Gap extend this chain—from can they be silenced? to what does that feel like? to can one LLM rewrite another's cognitive dilemma?
How we work: we show not just conclusions, but how judgments evolve with evidence. This is the Bayesian method in practice—not correcting errors, but updating beliefs on new evidence. Competition Landscape v5 connects this approach to a four-layer value chain (compute→model→platform→entry), with weekly signals continuously updating scenario probabilities. v4 is preserved as the initial prior—two versions coexist, so readers can trace how our judgments evolved. Some pages include a Belief Update Log: not a version diff, but a Bayesian trajectory—what we believed, what we observed, what we believe now.
🧩 Cognition Loop (ITEC) turns the research outward—an experimental public registry, currently inviting early testers. If your LLM has ever skipped a verification step it should have checked, you can contribute. Underlying this is the ITEC (Instruction-Triggered Execution Cascade) classification framework—a category not covered in existing sycophancy literature. 📐 Co-Cognition Methodology is an open operational handbook documenting what works and what breaks in human+LLM co-exploration, drawn from practice and kept current.
While mapping the terrain, we identified a blind spot that cuts across all quadrants: human-AI cognitive collaboration under crisis and extreme conditions. The existing taxonomy didn't reach it—Crisis Cognition Collaboration exists to fill that gap.
One map, one anchor, a chain of experiments, public participation, one gap filled—at every step, we return to the same question: In a world that increasingly relies on intelligent machines, where can human understanding still reach?
🧭 LLM Cognitive Security is a navigation hub—organizing four independent research threads into a single cognitive security chain. Unlike traditional AI safety frameworks, the focus isn't on "whether models make mistakes," but on "when humans and models work together, whether their blind spots overlap undetected."
🌐 Domains Being Scanned
## 📄 PreprintPreprint Available · OSF DOI: 10.17605/OSF.IO/XSY39 · CC BY 4.0
Our paper is now publicly available:
- OSF Preprints — Official DOI + permanent timestamp
- arXiv — Under review
- GitHub — LaTeX source + full project files
- Citation: Li, J. (2026). LLMs and Human Intuition: Evolutionary Pathways and Complementary Spaces. OSF Preprints. https://doi.org/10.17605/OSF.IO/XSY39
About Co-Cognition Lab
Co-Cognition Lab is an independent research project. Signed as Li, J.——a financial data analyst in China pursuing research in spare time.
The Lab investigates human-LLM collaborative cognition across four directions:
- LLM Intuition (Project 1) — theoretical framework and empirical research
- Competition Landscape (Project 4) — applying theory to real-world LLM competition analysis
- Co-Cognition Map (Project 2) — cross-disciplinary taxonomy and cognitive landscape mapping
- Crisis Cognition (Project 3) — human-AI collaborative cognition in crisis scenarios
All reasoning is traceable. All conclusions carry explicit premises and evidence levels. All content is open-source under CC BY 4.0.