Blog

Generate more evenly distributed results across multiple runs with random prompt injection (RPI), a technique that involves randomizing part of the prompt to force increased variety in outcome. Learn how to use RPI with helpers in Prompt Wrangler to inject random numbers, sentiments, or other variables.

When using prompts in agent-style workflows, it's important to know when to stop. Adding a boolean property called "complete" to the JSON response can lead to errors, so instead, turn "complete" into a tool. This change leads to better performance and easier model fine-tuning.

Getting GPT to return valid structured JSON with the right format can be a bit tricky. These tips will help get structured JSON in the right format every time.

Prompt Wrangler is a new service that makes it easy to turn GPT prompts into structured APIs