Write: geo@qiyanword.cn
阅读中文
One hypothesis, three directions, four deliverables. What follows is what I am working on, how far it has gone, and which parts have no answer yet.
Efficient-market theory asks whether prices have already absorbed information. The generative hypothesis changes the starting point: price is the residue of interaction, and information itself is produced inside that interaction.
| Question | Efficient market | Generative market |
|---|---|---|
| What is a market | An information aggregator that tends toward equilibrium | A process that keeps taking temporary shape |
| What is price | The discounted result of information | The residue of the last interaction, and a prompt for the next |
| Source of uncertainty | Incomplete information, progressively removable | Structure itself changes, so old information expires |
| Research task | Estimate true value | Identify generative mechanisms and turning signals |
| Decision meaning | Optimize allocation, beat mispricing | Generate options, reset the problem and the rules |
Treat an industry as an ecosystem rather than a racetrack: firms, consumers, regulators and technical standards are each other's environment, and structures form and collapse at thresholds.
Because strategy is built on the causality of the previous ecosystem. When the ecosystem changes, yesterday's optimum becomes a burden, while the organization keeps optimising the old metric.
Write firms, consumers and regulators as agents with rules, so that niches, subsidy wars and standards competition emerge in simulation instead of being assumed.
Who sits where, which rule yields the rent, and which signals indicate the structure is switching tracks.
From predicting the market to generating opportunity: define a problem nobody has framed, then create options that can be tested. Decision quality is whether it still holds when you do not know.
When structure itself moves, more precise forecasts can reduce the robustness of a decision. What you need is not a better point estimate but a better order of action.
Do not list possible futures; let scenarios arise from the interaction rules of the agents. Each path carries trigger conditions and observable early signals.
A workflow usable in a meeting: problem definition, scenario generation, signal monitoring, review and calibration. With a whitepaper and templates.
Agents in open worlds cannot live on pre-training. They learn while acting, evolve with people, and let system-level cognition exceed the sum of its parts.
Large models are good at problems that are already well posed. The difficulty in real decisions is that the problem is wrong, vague, or not yet posed.
Build agents that rewrite their objective through sustained interaction with the environment, exposing their own assumptions through action rather than fitting offline data.
An agent architecture for open tasks, plus controlled comparisons against conventional forecasting models.
Each phase leaves reusable theory and tools for the next one, rather than a list of disconnected projects.
This position sits in the applied layer of complex-systems science, with "generation" as the unifying frame.
| Institution | Its strength | What is added here |
|---|---|---|
| Santa Fe Institute | Foundational work on complexity, emergence, evolution | Translation into decision tools firms and investors can use |
| MIT Sloan | Scientific, empirical approach to management | The generative paradigm beyond predict-and-optimize |
| DeepMind / OpenAI | General AI capability and model scale | Agents tested inside the open environment of business decisions |
| Strategy consulting | Mature method and organizational reach | No certainty on offer; scenarios and signals that can be overturned |
A theory is more useful published in a discussable form than withheld until it is "complete".
Replacing departmental ontology with functional ontology: five functions in dynamic coupling, a three-layer superstructure, five separated powers, and five testable propositions.
Information potential difference explains coordination failure and transaction costs; three mechanisms and five testable propositions, with exploratory evidence from an automobile chain and a chemical park.
No paper yet — only a language of analysis and one architecture figure. Published early so that it can be refuted.
Restating "edge and generation" as a problem that already has a name: the state is hidden, so you maintain a belief; when the belief moves, the policy resets. Four concepts — non-equilibrium forcing, weak signals, attention density, generative value — each anchored to a computable object.
Research collaboration, closed-door workshops, quarterly advisory — all starting from a real problem.