<aside> 🛤️ Before ChatGPT was officially equipped with web browsing capabilities, I created a project to extend its reach into real-time internet data retrieval. This endeavor aimed to enhance ChatGPT's conversational responses by providing it with the ability to fetch and utilize live web content. I later learned this technique is called retrieval augmented generation (RAG).


DALL·E 2024-01-28 14.09.55 - A visual representation of a personal project integrating ChatGPT with real-time web browsing capabilities, before such features were officially avail.png

Project Background

The project was born out of curiosity to experiment with AI's potential to interact with the vast resources available on the internet. The goal was to enable ChatGPT to not just rely on its pre-trained knowledge base but also to access and incorporate up-to-date information from the web into its responses.

Implementation Overview

The project involved developing a Python-based system that acts as a bridge between ChatGPT and the internet and provides an interface with which the user can interact. Here's a brief look at how it works:

  1. Prompting GPT 3 with usage instructions: The first step to the program prompts GPT 3 with the starting text below which instructs the AI to only respond to questions with Search: <search query> or Answer: <the answer>, and that it will be returned results from Google in response to its Search: query. In response to a user question, GPT 3 will make its decision if it needs to search online to find the answer, and will respond as appropriate.
  2. Detecting and performing searches: Because of GPT 3’s usage instructions on how to format searches queries, my program can easily detect them, and then use the requests library to conduct a Google search: "<https://www.google.com/search?q=>" + query.replace(" ", "+")
  3. Content processing: The retrieved content is processed to strip away non-essential elements, using BeautifulSoup, to present clean and relevant text to ChatGPT.
  4. Feeding information to GPT 3: The cleaned-up content is then fed into GPT 3, enabling it to formulate responses with the latest information available online.

Reflections and Learnings

While this project was a personal exploration rather than a groundbreaking innovation, it provided valuable insights into the complexities and possibilities of enhancing AI with live web data. It predated the official web browsing capabilities now present in ChatGPT, offering a glimpse into the potential such features could hold.

Moving Forward

The official introduction of web browsing to ChatGPT has since overtaken the need for such external integrations. However, the experience of working on this project has been instrumental in understanding the intricacies of AI and web interaction. It stands as a testament to the continuous learning journey in the field of AI and the iterative process of innovation.

This portfolio project write-up was written in part by ChatGPT.

<aside> <img src="/icons/home_blue.svg" alt="/icons/home_blue.svg" width="40px" /> Home


<aside> <img src="/icons/document_green.svg" alt="/icons/document_green.svg" width="40px" /> Resume


© Jesse Gilbert 2024

Powered by Fruition