Self-Refine: Iterative Refinement with Self-Feedback
Explore this paper's citation graph
Summary
Self-Refine is introduced, an approach for improving initial outputs from LLMs through iterative feedback and refinement that demonstrates that even state-of-the-art LLMs like GPT-4 can be further improved at test time using this simple, standalone approach.
- Type
- preprint
- Published
- 2023-03-30
- Cited by
- 4,484
- References
- 52
- Access
- Open access
- OpenAlex
- https://openalex.org/W4362508231
- Semantic Scholar
- https://api.semanticscholar.org/CorpusID:257900871
Keywords
Task (project management), Computer science, Reinforcement learning, Generator (circuit theory), Dialog box
References
- A Cognitive Process Theory of Writing
- Interval Estimation for a Binomial Proportion
- Character-level Convolutional Networks for Text Classification
- Guide Me: Interacting with Deep Networks
- Delete, Retrieve, Generate: a Simple Approach to Sentiment and Style Transfer
- Style Transfer Through Back-Translation
- Improving Natural Language Interaction with Robots Using Advice
- Teaching a black-box learner
- CommonGen: A Constrained Text Generation Challenge for Generative Commonsense Reasoning
- Unsupervised Evaluation of Interactive Dialog with DialoGPT
- Graph-based, Self-Supervised Program Repair from Diagnostic Feedback
- NL-EDIT: Correcting Semantic Parse Errors through Natural Language Interaction
- Think about it! Improving defeasible reasoning by first modeling the question scenario.
- Memory-assisted prompt editing to improve GPT-3 after deployment
- Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback
- Read, Revise, Repeat: A System Demonstration for Human-in-the-loop Iterative Text Revision
- Learning to Model Editing Processes
- Quark: Controllable Text Generation with Reinforced Unlearning
- Self-critiquing models for assisting human evaluators
- CodeRL: Mastering Code Generation through Pretrained Models and Deep Reinforcement Learning
Cited by
- Large Language Models can Implement Policy Iteration
- Reasoning with Language Model Prompting: A Survey
- Language Models can Solve Computer Tasks
- REFINER: Reasoning Feedback on Intermediate Representations
- Teaching Large Language Models to Self-Debug
- Chameleon: Plug-and-Play Compositional Reasoning with Large Language Models
- Learning to Plan by Updating Natural Language
- Improving Grounded Language Understanding in a Collaborative Environment by Interacting with Agents Through Help Feedback
- Divide and Prompt: Chain of Thought Prompting for Text-to-SQL
- Enhancing Chain-of-Thoughts Prompting with Iterative Bootstrapping in Large Language Models
- ChatLLM Network: More brains, More intelligence
- Answering Questions by Meta-Reasoning over Multiple Chains of Thought
- ICE-Score: Instructing Large Language Models to Evaluate Code
- A Brief Overview of ChatGPT: The History, Status Quo and Potential Future Development
- Bridging the Gap: A Survey on Integrating (Human) Feedback for Natural Language Generation
- Self-Evaluation Guided Beam Search for Reasoning
- "Oops, Did I Just Say That?" Testing and Repairing Unethical Suggestions of Large Language Models with Suggest-Critique-Reflect Process
- An automatically discovered chain-of-thought prompt generalizes to novel models and datasets
- Self-Chained Image-Language Model for Video Localization and Question Answering
- Learning to Simulate Natural Language Feedback for Interactive Semantic Parsing
Related papers
- Where to Go for the Holidays: Towards Mixed-Type Dialogs for Clarification of User Goals
- Dialog's finder files: What will you find? What will you miss?
- The Role of Intention in Maintaining Coherent Human-Computer Dialog: Two Case Studies
- Notes on Dialog and Cooperation
- Reinforcement Learning in Multi-Party Trading Dialog