Browse with Bing Functionality Comes to ChatGPT

by Clay Casper, Solutions Developer - Corporate Branding and Jake Nelson, Data Analyst at FortisureIT

TL;DR, as generated by ChatGPT-3.5: Large Language Models like ChatGPT have an inherent limitation in their access to information beyond their training dataset. To address this limitation, the Browse with Bing beta feature, released in September 2023, provides up-to-date data to ChatGPT users, potentially revolutionizing applications of these models.

Large Language Models, or LLMs, have been taking the information and tech industries by storm over the last year. With ChatGPT, BardAI, and similar tools becoming household names, various organizations have been focused on ironing out issues and developing solutions to the limitations inherent in these models.

One limitation within ChatGPT is it's limited access to information outside of the model's training dataset. As made clear by a disclaimer on OpenAI's website, ChatGPT was only trained on data up to 2021, and the AI wouldn't have access to information post-2021. One way to get around this is with the use of Retrieval Augmentation Generation (or RAG) architecture. Put simply, this is where you take a base LLM, like ChatGPT's titular GPT, and write some custom code to fetch up-to-date data; finally injecting that data into the prompt you send to GPT. The issue with RAG is that it often requires setup by the end-user. Luckily, it seems that OpenAI has once again done some of the heavy lifting for us with an alternative solution!

First released as a plugin on March 23, 2023, and then as an experimental beta feature, the Browse with Bing beta feature is now available to Pro and Enterprise ChatGPT users as of September 27, 2023. This feature aims to rectify the same issues that RAG targets: bringing in additional data that the model wasn't originally trained on. While we can't be completely sure if it uses the RAG model under the hood or if they are doing some other sort of magic, we do know that queries about post-2021 content are no longer an issue... at least with ChatGPT Pro & Enterprise.

This feature is still in beta and is available through the Pro and Enterprise plans on ChatGPT, which currently start from $20 per month. While definitely not a completely stable feature (oftentimes ChatGPT will simply fail halfway through its response or just refuse to respond entirely), the implications of this feature being built-in to the ChatGPT client are promising.

From troubleshooting bugs in code with up-to-date documentation to conversing with custom AI personalities, the use cases that can leverage the power of methods like RAG and Browse with Bing are endless, and the introduction of new features like Browse with Bing puts these possibilities right on the horizon. Unfortunately, these features are not available through the API version of GPT or through Azure's OpenAI wrappers as of the time the article is published, limiting the range of possibilities (for now).

?

In short, imagine an AI that has access to the internet: answering all your questions with up-to-date information that's constantly growing, changing, and integrating with a massive language model that communicates clearly and easily. It's an exciting, and somewhat scary, thought, and it's much closer to us now than it's ever been.

If you'd like to know more about how we can help with analytics and development in your organization, please reach out at the link below:

Contact Us at FortisureIT

If you'd like to learn more about how LLMs and other models function, please check out the links below:

Flyte | Getting Started with Large Language Models

Financial Times | Generative AI