ChatGPT NET PlugIn Series
- Intro to ChatGPT Plugins using NET
- Map route handler for manifest file: ai-plugin.json
- GetAll and Add, 2 API endpoints in a single PlugIn
- Working and Debugging with the PlugIn Dev Tools
Hi!
One of the key elements while debugging a ChatGPT plugin is to understand how ChatGPT will invoke my plugins endpoints.
When a user asks a plugin-related question, the model digs into the API specs and the manifest file for answers. Just like testing a variety of prompts, you gotta play around with different descriptions to find the magic formula!
The OpenAPI spec is like an info-packed treasure trove for the model, telling it all about your API’s functions and parameters. Use clear names and add some “description” fields to explain the truth. The model will read it all up and become an API master.
The “description_for_model” attribute is the key part to guide the model on how to use your plugin. ChatGPT understands natural language like a boss and follows instructions like a pro. So, a clear description of the plugin, about what your plugin does and how it should be used. It should be descriptive and objective. Start with “Plugin for …” and complete the amazing things your API can do!
I strongly advice to check the sections [Writing descriptions] and [Best practices] from the official documentation,
And the last section [Debugging] is also a key one.
We can enable the Plugin dev tools, from our user settings, to validate how the GPT model understands our plugin definition.
We can always open the plugin call combo to see request and response details of the API call.
This is super cool. I’ll continue updating the repo with the demo instructions, and a live session will be scheduled soon!
Happy coding!
Greetings
El Bruno
More posts in my blog ElBruno.com.
More info in https://beacons.ai/elbruno
Top comments (0)