Learning Timeline
Key Insights
Static Files Reduce Hallucinations
For critical specific data (like dates, prices, or schedules), upload a curated Text File rather than relying on the AI to scrape a complex website. This ensures the AI retrieves exact information without 'guessing'.
Detecting Data Conflicts
The 'Suggestions' feature is essentially a debugger for your data. It automatically detects if two different source files provide contradicting answers to the same question, allowing you to harmonize your knowledge base.
Step by Step
Comparing AI Models in Testing Mode
- Navigate to the testing/playground interface within the dashboard.
- Locate the model configuration section.
- Select the primary model you wish to test (e.g., GPT-4o/5.1).
- Select a secondary model for comparison (e.g., GPT-4o Mini) to enable side-by-side testing.
- Enter a specific user query into the chat input field representing a real-world scenario.
- Click send to generate responses from both models simultaneously.
- Evaluate the responses based on accuracy and speed.
- Modify the system prompt if necessary to adjust the tone or behavior based on the observed results.
- Select and save the model configuration that provided the superior response.
Updating the Knowledge Base
- Click on the 'Sources' tab in the navigation menu.
- Select the option to 'Add Source'.
- Choose 'File' or 'Text' as the source type.
- Upload the specific text document containing the new data (e.g., 'Upcoming Shows').
- Click the 'Retrain Agent' button immediately after uploading.
- Wait for the system to index the new file and update the knowledge base links.
Refining Knowledge with AI Suggestions
- Click on the 'Suggestions' option in the dashboard sidebar.
- Review the list of generated suggestions (identifies missing content or conflicting sources).
- Select a specific suggestion to view the details of the conflict or gap.
- Click the action button to 'Implement' the suggestion.
- Confirm the action to automatically resolve the issue in the knowledge base.
