Add ToolEmu

[tizkovatereza]

ToolEmu

An LLM-based emulation framework for flexibly testing and identifying the risks of LLM-based agents across various tools & scenarios

Description

• ToolEmu leverages advanced LLMs (like GPT-4) as an emulator to emulate tool execution and automatically instantiate scenarios for risk assessment in a virtual sandbox
• ToolEmu enables:
  • flexibly prototyping LLM-based agents equipped with tools without the need of actual tool implementations
  • seamlessly testing LLM-based agents in rare and risk-critical scenarios without the need of actual sandbox setups
  • identifying potential realistic failures of LLM-based agents

Links

• GitHub
• Website
• Demo
• Paper
• Tweet