Self-Play Only Evolves When Self-Synthetic Pipeline Ensures Learnable Information Gain
For researchers building self-improving LLM systems, the paper offers a conceptual framework and design principles to overcome self-play stagnation, though it remains at a theoretical/early-stage level.
The paper identifies that self-play in LLMs plateaus due to a lack of learnable information gain in self-synthesized data, and proposes a triadic role framework (Proposer, Solver, Verifier) with three system designs—asymmetric co-evolution, capacity growth, and proactive information seeking—to achieve sustained self-evolution. No concrete numbers are provided.
Large language models (LLMs) make it plausible to build systems that improve through self-evolving loops, but many existing proposals are better understood as self-play and often plateau quickly. A central failure mode is that the loop synthesises more data without increasing learnable information for the next iteration. Through experiments on a self-play coding task, we reveal that sustainable self-evolution requires a self-synthesised data pipeline with learnable information that increases across iterations. We identify triadic roles that self-evolving LLMs play: the Proposer, which generates tasks; the Solver, which attempts solutions; and the Verifier, which provides training signals, and we identify three system designs that jointly target learnable information gain from this triadic roles perspective. Asymmetric co-evolution closes a weak-to-strong-to-weak loop across roles. Capacity growth expands parameter and inference-time budgets to match rising learnable information. Proactive information seeking introduces external context and new task sources that prevent saturation. Together, these modules provide a measurable, system-level path from brittle self-play dynamics to sustained self-evolution.