REFINER: Reasoning Feedback on Intermediate Representations
This addresses the issue of unreliable intermediate deductions in reasoning tasks for users of language models, representing an incremental improvement by integrating feedback mechanisms.
The paper tackles the problem of language models generating incorrect intermediate reasoning steps by introducing REFINER, a framework that finetunes LMs to generate reasoning steps while interacting with a critic model providing automated feedback, resulting in significant improvements over baseline LMs on three reasoning tasks and enhancing reasoning in GPT-3.5/ChatGPT without finetuning.
Language models (LMs) have recently shown remarkable performance on reasoning tasks by explicitly generating intermediate inferences, e.g., chain-of-thought prompting. However, these intermediate inference steps may be inappropriate deductions from the initial context and lead to incorrect final predictions. Here we introduce REFINER, a framework for finetuning LMs to explicitly generate intermediate reasoning steps while interacting with a critic model that provides automated feedback on the reasoning. Specifically, the critic provides structured feedback that the reasoning LM uses to iteratively improve its intermediate arguments. Empirical evaluations of REFINER on three diverse reasoning tasks show significant improvements over baseline LMs of comparable scale. Furthermore, when using GPT-3.5 or ChatGPT as the reasoner, the trained critic significantly improves reasoning without finetuning the reasoner. Finally, our critic model is trained without expensive human-in-the-loop data but can be substituted with humans at inference time.