How to Add Web Browsing Functionality to a Bot
Continuing the topic of plugins and the fact that the version with web search functionality is available only to a small number of testers through Alpha access.
As you know, to understand the technical details of GPT, I created the bot @rvnikita_public, which is essentially an extended version of GPT + Dall-e. It can hold a conversation, generate images, and has many additional features (for example, with the command /remember, you can set a memory that the bot will remember with each request and will never forget. For me, this is “`Act like an English teacher who tries to make users' English better, more native, and without mistakes. Always analyze user requests for grammar and linguistic mistakes. If you see any mistakes, include the correct version, information about the mistakes, and how to fix them AT THE END of your reply, separated by a line. Don't copy the user's input, as they know what they wrote.`” - which has allowed me to discover and correct a huge number of minor mistakes in my English.
But this post is not about that - it’s about my thoughts on how to add web browsing functionality to the bot. This weekend, I added the ability to search for information on the internet if the GPT model does not know it. It doesn't always work super smoothly yet, but the results are already impressive.
To add search results, I use the `Custom Search JSON API` and the Python library `googleapiclient`.
Globally, I see three approaches to adding information to GPT:
