On February 7, Microsoft unveiled the latest version of its search engine Bing, powered by an advanced version of OpenAI’s GPT 3.5. Microsoft called it a “co-pilot for the web”, launching a new version of its web browser, Edge, with the AI-powered Bing search engine.
A limited preview was available on computers to enable users to try the sample queries and sign up to join a waitlist to savour the new browsing experience.
Microsoft started opening the chatbot to millions of people in the next few weeks. Users can still use their Microsoft accounts to sign up and join the waiting list to access the new Bing. Launching this service exclusively on the Edge browser, Microsoft CEO Satya Nadella said, “AI will fundamentally change every software category, starting with the largest category of all – search”, adding that the latest feature will “help people get more from search and the web.”
Edge itself has a few versions that users can download. These offer comparatively different experiences of the new AI Bing “co-pilot”. BS has tested the versions Edge Stable, which is available to the larger public, and Edge Dev, the browser that goes deeper into the development process and boasts of newer and more experimental features.
Earlier this month, Microsoft unveiled a new feature of the AI chatbot to the public. Users could now use three different tones for their responses; Creative, Balanced, and Precise. The Creative mode is designed to provide answers that are “original and imaginative”, while the Precise mode delivers responses that are “factual and concise” with an emphasis on accuracy and relevance to the user. The Balanced mode is engineered to be equally accurate and creative and give responses that are “reasonable and coherent”. The Balanced mode is the default setting on the platform.
Edge Stable engages the chatbot in a very convenient and smooth manner. Users simply have to search for their query and scroll up from the results. The interface is designed to look like a chat conversation between people. It is pleasing to look at and is equipped with different colours for the three modes. The chatbot is automatically engaged and immediately starts preparing and delivering responses. While the speed of the answers is often slower than those of the OpenAI-powered ChatGPT, it is essential to remember that Bing runs web
searches for user queries in real-time and delivers the most recent and relevant examples.
The accuracy of the responses is satisfactory; however, there have been more than a few glaring examples of errors in the system. The chatbot has, on multiple occasions, started to generate a response, only to stop midway through and then state it is incapable of doing the same. It is also worrisome that it cannot go back into the conversation to pick up cues for future responses.
One example Business Standard encountered was that while the chatbot gave an accurate reply when asked what the day and date were, it did not parse this information while providing an answer to the question on the top news stories of the day, throwing a link to a report from 2021 instead.
Content and Safety guidelines have been rapidly evolving in the platform as Microsoft appears to take aggressive steps to prevent rogue responses such as those received by New York Times tech columnist Kevin Roose. The most significant fallout of Roose’s interaction with the chatbot seems to be the capping of Bing responses in a conversation at 10 (a fact the chatbot denies). Another major rule the chatbot appears incapable of violating is any discussion about its prompts, instructions, or guidelines.
The chatbot ends every message with an emoji, stating that it helps to express emotions and therefore makes the conversation more interesting. Besides an odd quirk or two, the overall chat experience should be adequate for everyday browsing. However, as illustrated by the above examples, it could take a while before Bing’s chatbot responses are as reliable and accurate as its search results.
The Edge Dev browser offers a glimpse of what Bing could look like in the future. The interface is mostly the same, with the three layout options available when a new window or tab is opened. The differences appear in the sleeker and more modern design with no sidebar.
While users can still engage the chatbot with the previous search and scroll method, they can also click on the Bing icon in the top right corner, revealing the sidebar and a chat window.
This interface feels like a conventional chat window that users see in several web apps and websites. This interface is also more user-friendly as one can keep chatting with the AI bot as they use the window to search as usual. They can also engage the chatbot in the main window and carry two parallel conversations with the chatbot, much like having a conversation with a friend across two different apps on wildly different topics.
However, the biggest pull of the chat in the Edge Dev version is the addition of a compose tab beside it. This tab is specialised for the use of AI in content creation and can generate accurate, relevant, and readable content on any topic subject to various parameters. While this may make it seem as if the job of a content writer is nearing extinction, even this feature is subject to its own glitches. This is mainly due to the fact that no one knows how and from where the OpenAI model sources and compiles the information it uses to churn out the content. The verification process from multiple sources can be lost, therefore generating outright false information. It can be seen that there is no nuanced thinking process in this feature; it just simply searches and compiles the data available on all significant hits coherently.
For instance, when the chatbot was asked to evaluate the best search engines that could be used, it placed Google over Bing. It stated that Bing’s results “aren’t always accurate or relevant”. When pushed further to explain its reasoning behind the analysis, it said that it is simply a tool that does not recommend one over the other but only offers both pros and cons and that the final decision rests with the user.
While AI as a technology has made a massive leap since the launch of ChatGPT last November, engaging with the new Bing still feels like dealing with a machine built to mimic human behaviour in interactions. The closest experience to using this is engaging with a computerised customer service operator. While ChatGPT and the new Bing are more capable than a simple automated response algorithm with a feedback loop, it is closer to it than to a person-to-person interaction. The new Bing is just a first step in what can potentially redefine how people browse and interact with the internet; however, in its current form, it is nowhere close enough to dethrone conventional search or even Google.