Artificial Intelligence, Center For Practice Management, Microsoft Office, Research

Bing Chat and Bard: The Basics of AI Chat Search Engines

Microsoft’s Bing and Google’s Bard are two chat-based search engines powered by artificial intelligence. What can they do? Let’s run through the basics, including how you can get access to these tools.

Google’s Bard

If you haven’t already, you can sign up to get on the waitlist to try Bard. Bard is listed as an experiment, and readily admits to having limitations and learns from feedback. To get on the waitlist you need a personal Google account. Google Workspace accounts cannot apply.

Google’s Bard is a conversational artificial intelligence model capable of dialog, built on Google’s LaMDA (Language Model for Dialogue Applications). Be aware that Google collects conversations you have with Bard, along with your IP address, feedback, and usage information. A subset of conversations is sent to trained reviewers and kept for three years, with automation tools to remove personally identifiable information. Google asks that you please not include information that could identify you or others in Bard conversations. Once you have access to Bard you can go to https://myactivity.google.com/product/bard and toggle off storing your activity. You can also delete your activity from this page.

Bard’s help screen suggested uses include brainstorming ideas, getting summaries, or creating first drafts of outlines, emails, and more. It suggests users should not rely on it for medical, legal, financial, or other professional advice. Bard’s data, unlike ChatGPT, comes from the internet.

Once you have entered a prompt, you can revise it by clicking on the pencil icon in the upper right corner. You can peruse three response drafts. When prompted to “draft a packing list for a business trip to Portland Oregon in July for a conference” Bard suggests taking business casual tops and dress pants/skirts, several pairs of dress shoes and one comfortable pair, a swimsuit, an umbrella, etc. It included a toiletries list and many other items.

The response is good, as good as the prompt let it be. For example, it doesn’t know where you are coming from, thus suggests an optional passport. It doesn’t know how you identify so the clothing suggestions are generic.

By refining the prompt to “draft a packing list for a female executive traveling for a business trip to Portland Oregon in July for a 3-day conference” the list became more specific and included business suits, down to 2 pairs of shoes (one comfortable, one dress), it added pajamas, some pain reliever and antacids to the toiletries list, and sleep headphones and an eye mask to the “other items”. It also added you may not need all the toiletries and personal items if you are staying in a hotel.

Interestingly, the more personal the prompt, the more personal the response. By further refining the prompt “draft a packing list for a female executive traveling for a business trip to Portland Oregon in July for a 3-day conference staying at the Portland Marriott traveling from Chicago” the AI suggested a work dress, a pair of flats (and identified Rothy’s as its personal favorite), and an optional pair of heels. It added a few other sundries. It also added a brief itinerary, a budget ($2500 for airfare, hotel, conference registration, food, and ground transportation) plus suggestions for travel contingencies. It is also linked to sources.

At the end of each response there was a link to “Google” your prompt to switch to the traditional search engine.

In a different test, the prompt was “write a biography for “catherine sanders reach”. Bard responded “I do not have enough information about that person to help with your request. I am a large language model, and I am able to communicate and generate human-like text in response to a wide range of prompts and questions, but my knowledge about this person is limited. Is there anything else I can do to help you with this request”? Interestingly there was no “Google This” link. Attempts to refine the query and rephrase it were met with the same response. Similarly, Bard supplied the same response to “who is Albert Einstein”. However, the query “write a brief biography for Albert Einstein” provided a short draft with sources and the choice to “Google it”.  The sources included a link to Walmart (a book description), Github, a site that gave a 404 error, and two Wikipedia pages.

There is a character limit in a prompt. When prompted to summarize a 10,000-character article it would only accept about 3,000 characters, although it summarized the snip well. However, by changing the prompt to “Summarize the article at https://www.ncbar.org/nc-lawyer/2022-11/three-reasons-lawyers-use-bcc-and-why-to-stop/”, Bard created a summary that started strong, but then it added some of its own “thoughts”, suggesting that “in addition to the risks mentioned in the article, there are a few other reasons lawyers should stop using BCC, including BCC can be a sign of poor communication skills”. This summary statement does not accurately represent any content in the article.

Bing Chat

Bing’s AI chatbot has a waitlist as well. The “new Bing” has the chatbot incorporated into the search engine. You may be logged in to a personal Microsoft account. The chatbot is also available in the Edge browser, in the Edge Mobile browser, and in the Windows 11 search bar. The new Bing screen prompts “Ask me anything”, up to 1000 characters.  You can run a regular web search, or toggle Chat.

Unlike Bard, you can’t see your previous chat logs, rather the chat and search query responses are stored in your search history. You can delete individual queries and clear all of your search history.

Chat in Bing is developed on code from their partner, OpenAI. OpenAI powers ChatGPT. Bing, owned by Microsoft, acknowledges the same foibles as Google. You can read their approach to responsible AI.

Bing Chat has a few different elements that contrast to Bard. First, the user chooses a conversational style. You can be “more creative”, “more balanced”, or “more precise”. In chat you can add 2000 characters, so an increase from the regular Bing search. Instead of editing your prompt, you can continue to “converse” with Bing Chat by adding qualifying questions until you choose “New Topic”.

Asking the same series of questions with refinements used to query Bard, Bing offers a brief and very generic packing list, and suggests follow up questions like “what is the weather in Portland in July” and “what are some good restaurants in Portland” when using the “more balanced” option. Applying the “more precise” option the same list for the query appears, but with links to “learn more” and some information about the average weather in Portland in July. It also checks to see if you have follow-up questions. By applying “more creative” to the same query, Bing said “Nice, this will be fun. What’s on your mind”.  It provided essentially the same list and wished me a good journey.

With Bing in creative mode, further exploration with more keywords did not yield different results. In “more balanced” the results were more refined as the query was more specific. In “more precise” mode the results shifted to more information about luggage choices. Interestingly, in a link to learn more, the content on Latterly.org’s website is evidently where Bard got its “preference” for Rothy’s shoes.

The results for each type of search (creative, balanced, or precise) are color coded so it is easy to distinguish which mode is used.

In the More Precise mode, Bing Chat generated a (mostly) factual and long response for the query “write a biography for catherine sanders reach”. In More Balanced the response was cut from 3 paragraphs to 3 sentences. In “More Creative” mode the biography was long, again 3 paragraphs, and unlike the other two results seemed to generate text not found in the hyperlinked bio where the text was taken from. For example, “Catherine Sanders Reach is a remarkable person who has made a significant impact on the legal profession through her work, passion, and dedication”. This sentence seems to have been generated by the AI, though much of the content in the results were simply synthesized text from the web. Re-running this prompt in different modes, using the main Bing Chat page and the Edge Browser plugin, also resulted in different versions of the biography every time.

When asked “Summarize the article at https://www.ncbar.org/nc-lawyer/2022-11/three-reasons-lawyers-use-bcc-and-why-to-stop/”, Bing Chat in More Balanced mode provided a clear and accurate three paragraph summary. In the More Precise mode the answer was shorter, but still correct. The More Creative mode got some of it wrong, and it was also a little confusing.

Bing Chat has a feature distinct from Google’s Bard. In the More Creative mode it will generate images. Image creation is powered by DALL-E.  Image Creator keeps copies of the images you created. There is no waitlist to use Image Creator, nor any costs, though you need to have a Microsoft account. It has restrictions so you cannot misuse certain images, like celebrities, public figures, and organizations.

Conclusion

While these tools are experimental, Bing and Google dominate the search space. Soon searches will be answered by an artificial intelligence combing the web and deciding what information to provide. It will be harder to assess the sources and determine the accuracy and timeliness. Like search engines, users will need to learn how to structure searches to effectively gather information. Because artificial intelligence can also synthesize or generate text, we will need to develop new skills to avoid the risk of relying on and propagating misinformation. Bing Chat and Bard are helpful for starting a draft of an email, letter, blog post, or ideas to give you some starting language but remember to fact check it!