It started with me trying to get consistent json output but then evolved to something I can use to get consistent outputs structured in the way I need with, which is essential to consume responses in software. I seem to get json consistent 100% of the time when I use the API but when I use the UI I occasionally some response data not in the json.
Unfortunate name/acronym (GPL will conflict with at least one well-established thing). Also - it’s a programming language to tell AIs what to do? I thought most of the point was that you can tell them what to do in natural language?
You can def tell AI what to do but the point is to have them do and reply to you in a way that's structured and expected, such that you can have your software process their responses and apply it according to your logic! With the least amount of words possible! The Human Response field is for people-facing responses (where you generally need more words) and the system response field is for responses to be used directly by your software.
Thank you for the suggestion! JSON Schema and a token corraler are definitely powerful tools for enforcing structure and validation. However, the goal of GPL is to provide a lightweight, easy-to-use way for users to structure their prompts without requiring them to install additional software or external dependencies.
While JSON Schema offers great flexibility, I'm focused on a format that works out of the box. I want to keep the process simple and frictionless, so introducing a token corraler with additional steps wouldn’t align with that goal.
That said, I’m open to incorporating JSON Schema as an optional feature for advanced users who want more control over their prompts. My goal is to keep installation as simple as copying and pasting the text at the beginning of the system prompt.
