Winner-Takes-All CLD
Open with full controls (sample buttons, file loader) Open the diagram in a clean view
This diagram is rendered by the shared CLD Viewer MicroSim, loading the file winner-takes-all-cld.json. The viewer is a port of the systems-thinking course CLD viewer.
Drag any node to rearrange the layout. Click a node, edge, or R/B marker for details. The full discussion is in the Winner Takes All article.
Loops
| ID | Type | Name | Role |
|---|---|---|---|
| R1 | Reinforcing | Recursive Self-Improvement | The core "winner-takes-all" hypothesis: better models build better models. |
| R2 | Reinforcing | Autonomous Research | Supercritical loop — closes inside the lab and decouples from human R&D pace. |
| R3 | Reinforcing | Capital → Compute → Capability | Where strategic investment (e.g. Google's $40B in Anthropic) plugs in. |
| R4 | Reinforcing | Data Flywheel | Proprietary interaction data competitors can't see — the lock-in mechanism. |
| B1 | Balancing | Compute Constraint | Compute scarcity throttles everyone, including the leader. |
| B2 | Balancing | Evaluation Bottleneck | Harder-to-evaluate models slow the rate of reliable progress. |
| B3 | Balancing | Diffusion / Fast-Follow | Visibility and talent mobility erode any leader's advantage. |
| B4 | Balancing | Cost-Performance Friction | High inference cost suppresses adoption and weakens the data flywheel. |