From RLVR to RLSVR: Task Transformation Induces Self-Verifiable Rewards for Open-Ended LLM Self-Improvement
Qinsi Wang · Jing Shi · Huazheng Wang · Kun Wan · Yiran Wu · Bo Liu · Qingyun Wu · Hai Helen Li · Yiran Chen · Handong Zhao · Wentian Zhao
Self-supervised learning, but for RLVR: we turn open-ended tasks into proxy environments whose rules make rewards verifiable. Our instantiation, SpyRL, is a "Who Is the Spy?" self-play game enabling LLM self-improvement on non-verifiable tasks.
摘要
Reinforcement Learning with Verifiable Rewards (RLVR) has driven recent progress in reasoning-oriented large language models (LLMs) by enabling large-scale optimization. However, its applicability remains largely limited to domains such as mathematics and coding, where correctness can be deterministically verifiable. Open-ended tasks instead often rely on human preferences, reward models, or LLM-based judges, introducing evaluation bias, judge capability bottlenecks, and additional inference costs. Drawing on the principle of self-supervised learning, which constructs pretext tasks to derive supervision from the data itself, we propose **Reinforcement Learning with Self-Verifiable Rewards (RLSVR)**, a task-transformation-based training paradigm for extending RLVR to open-ended tasks. RLSVR transforms open-ended tasks into verifiable proxy environments whose internal rules and interaction outcomes automatically generate reward signals. We instantiate RLSVR with **SpyRL**, a **S**elf-**P**la**Y** **R**einforcement **L**earning method inspired by social deduction game *Who Is the Spy?*. Agents receive asymmetric information, complete the same target task, and vote to identify a designated spy. Because the spy identity is predetermined, voting outcomes provide fully verifiable rewards, while successful identification remains closely related to output quality. Experiments on text summarization, creative writing, and mathematical reasoning show that SpyRL outperforms existing self-improvement methods on non-verifiable tasks and yields consistent gains on verifiable reasoning tasks. These results demonstrate that task transformation can extend scalable RLVR-based self-improvement beyond inherently verifiable domains. Models and code have been released at https://github.com/wangqinsi1/RLSVR/tree/SpyRL.