Agentic data science with the Data Agent Kit
cloud.google.com·7h ago
Data Science
Simon Willison's Blog·20d ago·5 min read
llm 0.36
TL;DR
The llm 0.36 update introduces model plugins that can specify single-turn prompt compatibility, preventing errors with conversation history. The first plugin to implement this is llm-typesafe.
✦ Why It Matters
Developers should implement llm-typesafe to ensure compatibility with single-turn prompt models today.
Key Takeaways
- Integrate llm-typesafe for safer model interactions.
- Avoid using conversation history with incompatible models.
- Check model compatibility before initiating sessions.
- Stay updated on plugin capabilities for better development.
Full Summary
The llm 0.36 update enhances the functionality of GPT-6 Luna by allowing model plugins to declare their compatibility with single-turn prompts. This means that models which do not support multi-turn conversations will raise an error, llm.ConversationNotSupported, if they receive any assistant or tool history.
Consequently, llm chat will reject these models before initiating a session, streamlining user interactions. The first plugin to utilize this feature is llm-typesafe, which ensures that developers can create safer and more predictable interactions with the model.
This update aims to reduce confusion and improve the overall user experience when working with different models. By clearly defining the capabilities of each model, developers can better tailor their applications to meet specific needs.