Generalized Agent Iteration: One Formal Framework for Iterative Policy Improvement and Recursive Self-Improvement
The paper tries to put ordinary policy iteration and self-improving agents on the same formal map.
GAI treats an agent as a set of modifiable components and frames learning as repeated evaluation and improvement. Its two key distinctions are whether the improvement mechanism sits inside the agent, and whether the evaluation standard is external or self-grounded. The authors use those axes to classify systems as anchored, drifting, or fully self-referential, and to state RSI failure modes more precisely. HF Daily Papers' note
GAI treats an agent as a set of modifiable components and frames learning as repeated evaluation and improvement. Its two key distinctions are whether the improvement mechanism sits inside the agent, and whether the evaluation standard is external or self-grounded. The authors use those axes to classify systems as anchored, drifting, or fully self-referential, and to state RSI failure modes more precisely. HF Daily Papers' note
score 4