loop de loop
· Zi Wang · 1 min read
R0 Output Recursion
├─ Scope The answer
└─ Action Generate → Critique → Revise
R1 State Recursion
├─ Scope Memory · Context · Skills
└─ Action Store lessons → Retrieve them next time
R2 Policy Recursion
├─ Scope Weights · Prompts · Tools · Harness · Source code
└─ Action Rewrite the agent’s own orchestration
R3 Evaluator Recursion
├─ Scope Tests · Judges · Reward models · Rubrics
└─ Action Improve what counts as “better”
R4 Representational Recursion
├─ Scope Concepts · Abstractions · Ontologies · Search spaces
└─ Action Invent a new language in which the problem becomes solvable
R5 Substrate Recursion
├─ Scope Training infrastructure · Chips · Robots · Institutions
└─ Action Improve the machinery producing future AI