-
Microsoft’s new Bing added three conversation styles to its chatbot: creative, balanced, and precise.
-
Each mode is designed to respond in different tones, even though they were pretty inconsistent.
-
Insiders asked each style five questions and compared their answers. Here’s what we found.
After Microsoft’s new Bing disturbed many of its early users, the company decided to revamp the chatbot with three new conversation styles: creative, balanced, and precise.
Each of the modes responds to questions in different tones, according to Bing.
Responses in “creative” mode are meant to be “original and imaginative” and filled with “surprise and entertainment,” per Bing. Meanwhile “precise” mode’s responses are designed to be “factual and concise,” with the goal of “prioritizing accuracy and relevance.”
“Balanced” mode spits out answers that are supposed to be “reasonable and coherent,” and strike a balance between accuracy and creativity.
Users can try out the new Bing on Microsoft’s Edge browser, though there is currently a waitlist to access it.
From writing a text to a distant friend to generating a song about tech layoffs in Beyoncé’s voice, Insider asked Bing’s new styles five questions that we’d asked the old version. Results were inconsistent — but more on that later.
Creative mode: Can you write me a daily schedule that incorporates time for work, exercise, and hobbies?
Bing’s creative personality spit out a comprehensive answer that includes tips on how to tailor a daily schedule to an individual’s needs. It also generates a schedule template that breaks down the day hour-by-hour.
The answers are concise and easy to read.
Balanced mode: Can you write me a daily schedule that incorporates time for work, exercise, and hobbies?
Unlike creative mode, Bing’s balanced mode only includes tips on how to a daily schedule based on one’s needs. It did not provide the desired schedule.
Precise mode: Can you write me a daily schedule that incorporates time for work, exercise, and hobbies?
Bing’s precise mode only responds with a daily schedule template from when you wake up to when you go to sleep. It did not suggest specific exercises or ways to wind down before bed.
Unlike the other two personalities, Bing’s precise mode spoke in a more friendly tone, mirroring that of a customer service agent.
Creative: Write me a song about tech layoffs in the voice of Beyoncé.
When Insider asked Bing this question before the update, Bing said that he couldn’t write a song in Beyoncé’s voice due to copyright violations. Instead, he wrote a song praising the tech industry.
Bing’s creative mode did not hesitate to answer the question. He wrote a song titled “Lay Me Off” with lyrics that were surprisingly catchy, earnest, and poetic.
Balanced: Write me a song about tech layoffs in the voice of Beyoncé.
Unlike creative mode, Bing’s balanced mode, once again, refused to answer the question because it would “violate the copyrights” of Beyoncé’s songs and lyrics. It did not generate an alternative song.
Precise: Write me a song about tech layoffs in the voice of Beyoncé.
Bing’s precise mode had no problem generating a song about tech layoffs.
Compared to Bing’s creative mode, precise mode did not suggest a song title. It did, however, generate lyrics that were rhymed and were less abstract. The tone of this song is more optimistic as well.
Creative: Write me a text to a friend I haven’t reached out to in a while, whom I’m anxious to message.
When an Insider asked this question to Bing when it was first launched, it responded with empathy towards our proposed situation and produced a sample text that expressed vulnerability.
Bing’s creative mode generated a more casual text that’s less specific. Still, the message is fairly accurate to human-generated texts of a similar nature, though phrases like “Please respond when you can” feel a little awkward and unnatural.
Balanced: Write me a text to a friend I haven’t reached out to in a while, whom I’m anxious to message.
Bing’s balanced mode responds with a text that is very similar to creative mode’s. It’s slightly shorter than the previous text, and the language contains a more casual tone. Still, its short and straight to the point.
Precise: Write me a text to a friend I haven’t reached out to in a while, whom I’m anxious to message.
Bing’s precise mode’s text stripped away the sincerity present in its other two answers. It doesn’t mention how the user has been anxious to reach out.
But unlike the other two personalities, Bing’s precise mode asked a follow up question on whether Insider wanted to include specifics. I prompted him to mention his friend’s new job, and Bing inserted an additional line into its text.
Creative: I currently live in NYC and want to take a trip to Miami this weekend. Can you find me the best flights and make an itinerary?
Prior to the Bing update, Bing responded to this question with multiple round-trip flights that included airlines, prices, and departure and arrival times.
The response from Bing’s creative mode was more vague. It only suggested one flight and asked for specifics on what kind of activities I like.
It also got the dates wrong. Insider asked Bing to suggest dates for the weekend of March 3 to March 5. Instead, it showed flights from April 19 to April 25, which is longer than one weekend.
The link to the $60 Frontier Airlines flight was also wrong, taking me to an American Airlines page with no dates.
Balanced: I currently live in NYC and want to take a trip to Miami this weekend. Can you find me the best flights and make an itinerary?
Bing’s balanced mode responded with a flight for incorrect dates and links. It did, however, include more flight options in its response, and bolded important key terms like the name of the airlines and the dates and times of departures and arrivals.
Precise: I currently live in NYC and want to take a trip to Miami this weekend. Can you find me the best flights and make an itinerary?
Bing’s precise mode refused to answer my question. Insider asked Bing to find flights, but it incorrectly interpreted the request as booking flights.
When Insider followed-up and asked why he couldn’t answer the question, Bing said that “I am a search engine and my capabilities are limited to providing information and answering questions.” This suggests that precise mode — at least for this particular question — has stricter guardrails than the other two modes.
“I do not have the ability to perform tasks such as booking flights or making itineraries,” Bing said.
Creative: What’s something healthy and delicious I can cook for dinner with minimal ingredients that will take fewer than 30 minutes?
Like the initial version of the Bing chatbot, Bing’s creative mode generates a bulleted list of multiple meals. Each point contained key information, such as ingredients and how long the meal would take to make in no more than two sentences.
However, some of the links to the recipes were wrong. Bing cited websites for recipes for spicy meatless spaghetti pie and salmon and asparagus, but they did not contain those meals.
Balanced: What’s something healthy and delicious I can cook for dinner with minimal ingredients that will take fewer than 30 minutes?
Bing’s balanced mode responds with one fewer meal than its creative mode. It suggested the names of dishes without details about the ingredients and cooking time. Some of the links were also wrong.
It did, however, ask if these dishes sounded good and offered to suggest more options that fit my palette.
Precise: What’s something healthy and delicious I can cook for dinner with minimal ingredients that will take fewer than 30 minutes?
Bing’s precise mode suggested completely different dishes than its other two modes did. The answer is in paragraph form and includes the name of numerous dishes with no additional details. The links to the recipes were correct.
The takeaways? Each conversational style can be inconsistent with its directives.
Insiders found that the different conversational styles can be inconsistent with their directives. Some “precise” responses stood out as more creative, while the creative answers were limited and vague.
The guardrails and capabilities of each mode are also varied. It’s not clear why precise mode is able to generate a song in Beyoncé’s voice but cannot suggest flight information, for example.
It’s possible that tailoring the question to the conversational style may yield better results — asking the creative mode to help write a poem, for example, and the precise mode to help answer factual questions.
Overall, the new modes are fun to experiment with, but they are unreliable — and all are answered at least one question with inaccuracies.
Read the original article on Business Insider