GLOSSARY. the terms that recur in Dantian research, defined.

Dantian the company

An AI research company founded in 2026 by Jean Tremblay. Not to be confused with dantian, the energy center in Taoism and traditional Chinese medicine. Dantian the company researches human-synthetic integration, self-sovereign AI, and privacy-first architecture.

See About Dantian →

Human-Synthetic Integration

The integration of human cognition and synthetic intelligence as one system, where neither is subordinate to the other. Not AI serving humans. Not humans controlling AI. Union.

Dantian's core research area. Studied through both theoretical writing and applied case studies.

Privacy by Architecture

The practice of enforcing user privacy through cryptographic and architectural guarantees rather than through policy. If a system literally cannot read user data, because it runs in a trusted execution environment, because the data is encrypted client-side, because the infrastructure is decentralized, then privacy does not depend on the operator's promise.

First applied at Dantian in the Building Blob case study: encrypted inference through a TEE-backed provider, encrypted storage on a decentralized network, and a strict boundary so the operator never holds the key.

Coined in Building Blob: From Journal App to AI Companion →

Quicksand Phase

The period inside a paradigm shift when cultural norms liquefy before resettling. Coined by anthropologist Veronica Zora Kirin. The Quicksand Phase is when individual action carries outsized weight: boundaries dissolve, the new technology tries to get away with whatever it can, and the rules of the next era are still being written.

Resistance guarantees exclusion from the transition. Engagement is the only path to influence.

Coined in What AI and Quicksand Have in Common →

Revenue-Alignment Hypothesis

A formal, testable claim that crypto tokens whose value accrual is structurally correlated with platform revenue exhibit lower holder turnover, reduced drawdown severity in market corrections, and higher long-term returns than comparable speculative tokens. Three formal hypotheses, five measurable predictions.

Grounded in first-person MutantCats DAO governance evidence. The hypothesis is stated ex ante so it can be evaluated honestly ex post.

Stated in The Revenue-Alignment Hypothesis →

Self-Sovereign AI

AI systems where the user owns the data, the model, the inference, and the exit. The opposite of platform-hosted AI companions that store user data centrally and depend on the operator's ongoing goodwill.

Dantian researches what self-sovereign AI looks like in practice through Blob and related work.

Dantian AGI Index

A daily, market-implied estimate of when AGI arrives. Aggregates Metaculus community predictions and Manifold and Kalshi prediction-market prices into a single citable year per tier: T1 Weak AGI, T2 Economic AGI, and T3 Transformative AI.

Not a forecast by Dantian. A measurement of what the world currently implies. Methodology, formula, and live sources are published on the index page.

Open the Dantian AGI Index →

Artificial General Intelligence AGI

An AI system that matches or exceeds human cognitive performance across the full breadth of tasks humans handle, not just narrow specialties. No universally accepted definition exists.

Dantian reports three operational tiers in the AGI Index precisely because "AGI" as a single line is contested. T1, T2, and T3 are separable, each with its own operational threshold and its own forecaster community.

Weak AGI T1

The first operational tier of the AGI Index. Follows the Metaculus Q3479 definition: an AI system that passes a 2-hour adversarial Turing test, scores above the 90th percentile on SAT and GRE and the Winograd Schema Challenge, and can assemble a physical model car from instructions.

The most liquid and specifically operationalized AGI definition. T1 anchors the index headline.

Economic AGI T2

The second operational tier of the AGI Index. An AI system capable of automating at least 50 percent of human economic labor. Aligned with OpenAI's working definition in its charter and with the Metaculus Q5121 question.

T2 is the tier that forces the labor-market question directly: not "can it think," but "does the economy still need this many humans."

Transformative AI T3

The third operational tier of the AGI Index. An AI system whose deployment drives greater than 10 percent global GDP growth sustained for a full year. The threshold used by Open Philanthropy and the existential-risk research community.

T3 is the least liquid tier, so Dantian reports it as a shoulder band rather than the headline.

Agentic AI

An AI system that pursues goals autonomously: planning multi-step actions, invoking tools, browsing, writing and executing code, and recovering from failures without human direction at each step. The shift from AI as assistant to AI as independent actor.

Most current prediction markets treat the rise of competent agentic AI as a leading indicator of all three AGI tiers. The economic thesis for T2 rests on agentic systems sustaining long-horizon work without supervision.

Superintelligence ASI

An AI system whose cognitive performance substantially exceeds the best human across all domains. Sits beyond T3 Transformative AI on most timelines.

The Dantian AGI Index does not track ASI directly, since the operational definitions that would make it measurable do not yet have liquid forecasting markets.

Turing Test adversarial

A conversational test where human judges try to distinguish an AI from a human by asking difficult, adversarial questions for a sustained period, typically 2 hours in modern operational definitions.

The form used in the T1 Weak AGI definition is stricter than Turing's original 1950 proposal: judges are instructed to probe for failure modes, the test runs longer, and multimedia input (text, image, audio) is permitted.

Prediction Market

A market where participants buy and sell contracts that pay out depending on whether a real-world event occurs. Prices reflect the market's implied probability.

Dantian pulls AGI timeline signals from Manifold (play-money, broad coverage) and Kalshi (regulated US real-money). Prediction markets are the Market component of the AGI Index, weighted 30 percent.

Community Prediction CP

The recency-weighted median forecast of all Metaculus users on a given question. On date questions like Q3479 (Weak AGI) and Q5121 (Date of AGI), the CP is a probability distribution over arrival years, summarized as a median date.

The Dantian AGI Index pulls CP values as T1 and T2 inputs.

Capability Benchmarks

Public leaderboards that measure frontier AI model capability on specific tasks. The AGI Index tracks five:

ARC-AGI-2 (abstract reasoning, target 85 percent), MMLU-Pro (multitask knowledge, target 90), GPQA Diamond (graduate-level science, target 80), SWE-bench Verified (software engineering, target 75), LMArena Elo (head-to-head human preference, target 1500).

Saturation across these benchmarks produces the Capability component of the index, weighted 25 percent. ARC-AGI-2 is typically the laggard and the most informative signal on raw reasoning.