JUNE 06, 2023
BY MARK FULTON
I’ve heard a lot of people confused or questioning the reason you can only have 3 plugins activated on ChatGPT.
It actually provides some insight into prompt engineering by showing the limitations of token counts.
According to OpenAI’s Plugin documentation, each plugin is allowed an 8000-character description that will assist the model in providing context to improve the prompt experience of for users.
So with 3 plugins combined that is a potential 24,000 characters or approximately 8000 tokens total which are included in the system context.
Then end-users like us are allowed up to 32,000 tokens.
So you can now see why the 3 plugin limitation is in place, namely because ChatGPT has to know how to use the plugin and provide instructions about the details of each of the three plugins.
Extracting the Context of Plugins
I experimented with extracting this system context from plugins with varying degrees of success.
How does the Diagr.am Plugin work?
Here’s a 🔗 ChatGPT thread showing some of the context provided to OpenAI by the developers of the Diagram plugin.
Want to Learn Prompt Engineering?
I’m teaching everything I know in The Ultimate AI Course. There’s a heavy focus on prompt engineering and developing AI tools with no code applications.