This morning I got access to ChatGPT's Browsing Model.

OpenAI's ChatGPT has long been known for its ability to generate human-like responses, but it was previously limited by a knowledge cutoff date. That has all changed now. Thanks to the introduction of ChatGPT's browsing model, this AI-powered chatbot has gained the ability to access and retrieve up-to-date information from the web.

Last week, as I described here, OpenAI announced a significant upgrade to ChatGPT, introducing a set of plugins as well support for web browsing. PC World explained how this update has made ChatGPT "a lot more powerful, overnight."

This new capability opens up a whole new world of possibilities for ChatGPT, enabling it to provide real-time answers and contextually relevant information. With this alpha release, ChatGPT is no longer stuck in the past and is ready to explore the present. In this blog post, I'll walk you through the key features, the capabilities I observed, and potential use cases of this new browsing model.

Overview of ChatGPT's Browsing Model

So, what makes this browsing model so significant? In essence, ChatGPT's browsing model enhances the chatbot's capabilities by giving it access to a simplified web browser tool. While traditional language models have a fixed knowledge base up to a certain cutoff date, ChatGPT with browsing can actively search the web, pull information from web pages, and even read articles or blog posts to provide users with the latest and most accurate answers. This means that whether you're looking for recent news, market data, or the latest trends in AI research, ChatGPT will be equipped to provide you with up-to-the-minute insights. (With regards to market data, see my caviat below.)

At the core of this advancement is the introduction of plugins, specifically designed tools that allow ChatGPT to interact with external sources of information. One highly anticipated plugin is the integration with the Bing Search API which empowers ChatGPT to run search queries, retrieve search results, and access a wealth of online information. The browser plugin essentially enables ChatGPT to navigate search results and extract relevant text from online content. OpenAI states, "With safety as a fundamental principle, these plugins are carefully designed to prevent misuse and promote ethical interactions."

As we delve deeper into the capabilities and use cases of ChatGPT's browsing model, let's take a closer look at how this bleeding-edge technology performed when faced with several different tests.

What the Browsing Model Can Do

As someone that has been following the AI space closely, I was curious to see how ChatGPT's browsing model would perform when tasked with retrieving specific information. For instance, when I asked, "What were the number 1 songs over the past 4 weeks?" the model initially struggled, but with slight modifications to the query, I finally did receive the desired answer.

Still didn't work.

Let's try again with a different prompt.

There we go. I guess my first two queries were too ambiguous. The specificity helped.

I'm also a bit of a car buff and over the past couple days I couldn't help notice more driverless Waymo cars here in my SF neighborhood. They've been testing with drivers for years but only recently have they been navigating our 'hood driverless.

So I tried a new query to see how well ChatGPT would find info on this. The query was "Which companies are currently operating self-driving cars in San Francisco? What are the latest developments in this area locally?"

Here is how Google's newly released AI chatbot, Bard, fared:

Not bad at all.

ChatGPT struggled on this one and got stuck:

ChatGPT proved to be a reliable copilot when looking up post-2021 facts. When I asked, "Who is the current prime minister of the U.K.?" I received an accurate answer without delay (see screenshot).

This query gave me an idea for a follow-up question. Would it have up-to-date information, and be aware of a meme that was circulating about a head of lettuce lasting longer than the previous British prime minister's reign?

I asked ChatGPT what my previous prompt may have to do with lettuce. It wasn't aware of any connection but asked if I had a specific topic or news article in mind. I hinted, "the previous prime minister."

At that point, it said it couldn't find any recent news articles that link Boris Johnson to lettuce. This was a surprising answer. I then asked who was prime minister between Johnson and Sunak, at which point it correctly named Liz Truss. My final query on this mini adventure was to ask, "what's the story with Liz Truss and lettuce?" which led to a spinning wheel which kept on going on and on. Perhaps ChatGPT was waiting for a fresh head of lettuce sitting somewhere to decompose before it would give me an answer. If that's the case, the joke's on me.

Back to the task at hand.

I was impressed by ChatGPT's ability to provide a two-pronged answer. When I inquired about the weather forecasts for San Francisco and New York City over the next three days, ChatGPT provided answers for both (see animation).

Similarly, when I asked for a recently published chocolate chip cookie recipe, it delivered. When I followed up with the question "How many calories are in this?" ChatGPT understood the context and provided the calorie count for the same recipe.

Current Limitations of the Browsing Model

Though ChatGPT's browsing model is powerful, it's not without limitations. For example, when I asked for Microsoft's current stock price, the initial response was outdated by several hours (it showed the price as of market close yesterday). I'm not sure if this is an issue with Bing's search results, or perhaps something that is currently built in to ChatGPT as a safeguard.

And although ChatGPT nailed the SF/NY weather answer, complex multi-pronged queries can be challenging for the browser model. When I asked for the "current temperature in the five largest U.S. cities", ChatGPT struggled to land at a response. It did first do a search for info on U.S. city population data, and then proceeded to search for weather info in each. Observing its interactions with the Bing Search API, I could see its efforts to piece together information from multiple sources. Ultimately it timed out while attempting to finish processing the result.

The browsing model is not without its weak points. But it is a remarkable step forward in AI-assisted browsing. As with any leading edge technology, there will be areas for improvement, and I think it's important to manage our expectations.

Real-World Use Cases

Businesses have already begun harnessing the power of ChatGPT by integrating the model into their apps and products via APIs. This integration is transforming how businesses interact with customers, provide support, and automate tasks. From customer service to content generation, ChatGPT's newly released browsing model will become a valuable tool for businesses seeking to leverage AI technologies (see examples of use cases).

Whether you're an experienced developer or a coding novice, ChatGPT's browsing model can be an invaluable resource for coding and troubleshooting. It can provide code examples, look up syntax and parameters, and help troubleshoot by offering solutions to common coding issues. Need to translate text? Just give ChatGPT your text, and it'll take care of it.

ChatGPT's browsing model is a content creator's dream come true. From generating blog posts and social media content to crafting engaging marketing materials, ChatGPT can help businesses and individuals create compelling content that resonates with their audience. That has been the case for several months now, and starting this week, it can help include (and name) recently published sources.

Looking Forward

I think the alpha release of ChatGPT's browsing model marks a significant milestone in the field of AI. Its ability to retrieve current information, answer questions, and engage in context-aware conversations opens up a world of possibilities. Of course, OpenAI's journey is not without its hurdles, and we observed areas where the model encountered limitations. Yet, the significance of this alpha release lies in its potential to evolve and improve.

Looking to the future, the potential applications of ChatGPT's browsing model extend far beyond what we've covered in this blog post. It's not just about answering questions—it's about revolutionizing the way we interact with information and technology.

As an advocate for AI innovation and the creator of www.theAInavigator.com, I'm excited to witness the evolution of ChatGPT and its contributions to various industries, from healthcare and finance to entertainment and education.

Additionally, I encourage you to try out ChatGPT's browsing model and experience its capabilities firsthand. As you do, consider sharing your thoughts, feedback, and observations with the broader AI community. Your contributions help shape the future of AI and foster a collaborative environment of learning and discovery.