Using LLMs to Speed Up Text Labeling in Label Genie
- Generative AI
This lesson walks through the generative AI features in H2O Label Genie's text labeling workflow, building on an earlier summarization example. It shows how to connect H2O GPT, or a custom self-hosted model that keeps data off external servers, and use prompt templates to generate a satisfaction score and highlight an issue for each review in a product review dataset. It demonstrates reviewing and approving the model's zero-shot predictions one by one, adjusting prompts as needed, then exporting the approved labels once the pass is complete.
Key takeaways
- How to connect H2O GPT or a self-hosted model to Label Genie so review data never leaves the environment
- Why customizing the prompt, for example asking for a satisfaction score, changes what the LLM extracts from a review
- How to review and approve zero-shot label predictions one at a time before exporting the final set
- What settings to enable first, such as zero-shot predictions and keyboard shortcuts, to speed up annotation
Start owning your AI