Learning to Reason from Feedback at Test-Time

Solving complex tasks in a single attempt is challenging for large language models (LLMs). Iterative interaction with the environment and feedback is often required to achieve success, making effective feedback utilization a critical topic. Existing approaches either struggle with length generalization or rely on naive retries without leveraging prior information. In this paper, we introduce FTTT, a novel paradigm that formulates feedback utilization as an optimization problem at test time. Additionally, we propose a learnable test-time optimizer, OpTune, to effectively exploit feedback. Experiments on two LLMs across four reasoning datasets demonstrate that FTTT and OpTune achieve superior scalability and performance.

@article{li2025_2502.15771,
  title={ Learning to Reason from Feedback at Test-Time },
  author={ Yanyang Li and Michael Lyu and Liwei Wang },
  journal={arXiv preprint arXiv:2502.15771},
  year={ 2025 }
}