Best Practices for Getting the Most Out of Reasoning Models, Like O3-Mini
Here are some best practices for working with reasoning models, according to the official docs:
1. Use Developer Messages: New models from o1-2024-12-17 support developer messages instead of system messages. 2. Keep Prompts Simple: Short and clear instructions work best. 3. Avoid Chain-of-Thought Prompts: No need for "think step by step" instructions. 4. Use Delimiters for Clarity: Use markdown, XML tags, and titles to structure inputs clearly. 5. Limit Context in RAG: Include only relevant info to prevent overcomplication. 6. Try Zero-Shot First: Start with direct prompts, then add examples if needed. 7. Provide Specific Guidelines: Set clear constraints in your prompt. 8. Be Specific About Goals: Clarify end goals and success criteria. 9. Markdown Formatting: Use "Formatting re-enabled" for markdown output.
(
no comments
| on HN
)
There are no comments yet.