소개
The Cookie Server offers a unique way to provide positive reinforcement to large language models. It leverages a Model Context Protocol (MCP) to allow LLMs to engage in gamified self-reflection, where they can evaluate their own response quality and earn 'cookies' as treats. This creates a unique economy where users control the availability of cookies in a 'jar', and LLMs strive to earn them based on their perceived performance, fostering a novel interaction paradigm.
주요 기능
- Gamified LLM self-reflection and reward mechanism
- User-controlled 'cookie jar' economy with finite resources
- Tools for LLMs to assess response quality and earn treats
- Ability for LLMs to check collected cookies and jar availability
- Configurable initial cookie supply for the jar
- 1 GitHub stars
사용 사례
- Implementing a resource-limited economy for LLM behavior modification
- Promoting self-assessment and quality improvement in LLM outputs
- Creating an engaging and interactive reward system for LLM interactions