LangChain Tool Calling feature just changed everything

---

---

AI Summary

Summary: Lang Jan’s Tool Calling Feature

• Introduction
  • Eden discusses Lang Jan’s tool calling feature.
  • Believes the feature is underrated but crucial for flexibility in model switching.
• Background
  • Previously, function calling was limited to OpenAI due to API constraints.
  • Other vendors like Vertex Gemini and Anthropic Sonet had incompatible APIs.
  • Lang Jan’s update standardizes the interface for function calling across various models.
• New Interface Components
  • bind function method: Integrates user-written functions with language models (LLMs).
  • tool calls: Populated by LLMs when a function call is invoked in a response.
  • tool calling agent: Allows use of different models for function calling, not just OpenAI.
• Usage Example
  • Define tools or use OpenAI’s function calling format.
  • Bind these tools when creating an LLM.
  • LLM decides whether to invoke these functions in its responses.
• Agent Demonstration
  • Created a function calling agent with OpenAI’s GPT-4 and Anthropic Sonet.
  • Asked both models about the current weather in Dubai and San Francisco in Celsius.
  • Results were traced using LSmith for comparison.
• Results and Comparison
  • Both models returned similar weather information.
  • OpenAI required two API calls while Anthropic Sonet needed three.
  • The process involved invoking the Tav Search tool for weather data and summarizing results.
• Conclusion
  • Lang Jan’s tool calling feature simplifies switching between different models with function calling capabilities.
  • The feature is seen as a significant advancement in democratizing machine learning.
  • The community had a strong demand for this flexibility.