noahshinn/reflexion
A NeurIPS 2023 framework enabling language agents to improve through verbal self-reflection and reinforcement learning.
Not currently ranked — collecting fresh signals.
star history
Reflexion implements a novel approach to language agent training where agents learn from verbal feedback rather than traditional RL gradients. The framework provides different agent types (ReAct, Chain-of-Thought) and reflexion strategies that allow agents to self-evaluate and refine their reasoning across multiple attempts. The implementation includes benchmarks like HotPotQA for evaluating multi-step reasoning capabilities.
Frequently asked
- What is noahshinn/reflexion?
- A NeurIPS 2023 framework enabling language agents to improve through verbal self-reflection and reinforcement learning.
- Is reflexion open source?
- Yes — noahshinn/reflexion is open source, released under the MIT license.
- What language is reflexion written in?
- noahshinn/reflexion is primarily written in Python.
- How popular is reflexion?
- noahshinn/reflexion has 3.2k stars on GitHub.
- Where can I find reflexion?
- noahshinn/reflexion is on GitHub at https://github.com/noahshinn/reflexion.