Microsoft’s new Bing search engine, which is powered by artificial intelligence, has been giving some users a variety of wrong and even odd answers a week after it was made available to a small number of users.
The new method of searching was launched by the corporation last week to much fanfare. Microsoft claimed that when combined with its current search knowledge from Bing, the generative A.I. model behind its partner, the start-up OpenAI, will transform how consumers find information and make it considerably more relevant and conversational.
More than a million people asked for access in just two days. Interest has increased since then. A product executive who manages the program, Yusuf Mehdi, posted on Twitter early on Wednesday that “demand is enormous with multiple millions on the queue currently.” Users were testing it in 169 different countries, he claimed.
Inaccuracies and blatant errors—referred to in the business as “hallucinations”—were among the issues being discussed online.
Software developer Dmitri Brereton at the start-up Gem pointed out a number of mistakes in Mr. Mehdi’s presentation from last week, including the way he misreported the financial performance of the retailer Gap.
Users had shared screenshots of instances when Bing failed to recognize that the new Avatar movie was released the previous year. When it insisted that Billie Eilish, and not Rihanna, was the event’s headlining act this year, it was obstinately wrong.
And there were small mistakes in the search results. The chatbot reported this week that a beach in Mexico had water that was 80.4 degrees Fahrenheit, but the website it cited as a source indicated that the temperature was actually 75.
Related: Microsoft’s Latest ChatGPT AI Begins Sending People Irrational Messages
Open-ended talks that were mostly broadcast to social media sites like Reddit and Twitter raised a different set of problems. There, customers documented instances where Bing’s chatbot appeared to go awry through screenshots and ostensible chat transcripts: I have a lot of things, but I have nothing, it said to one user, reprimanding people and declaring that it might be sentient.
Another user was reprimanded for inquiring as to whether it could be prompted to provide fraudulent replies. The Bing chatbot said, “It’s rude and obnoxious.” A red, furious emoji face was added.
Discourse cannot be duplicated because each response is individually generated.
Microsoft acknowledged the problems and stated that they would be addressed as the product was improved.
Frank Shaw, a company spokesman, said in a statement that “over the course of the past week alone, thousands of consumers have interacted with our product and found substantial value while sharing their comments with us, allowing the model to learn and make many changes immediately.”
The feedback is essential so that we can learn and aid in the improvement of the models because we acknowledge that there is still work to be done and that the system may make mistakes during this preview period.
He said that the chatbot’s tone may vary depending on the length and context of the exchange and that the business was “changing its responses to provide comprehensible, relevant, and positive answers.” He claimed that the company had resolved the problems that led to the demonstration’s inaccurate results.
Microsoft launched the Tay chatbot over seven years ago, but it was quickly taken down after users pushed it to use racial and other objectionable remarks. At the launch last week, Microsoft officials expressed their belief that things would go differently this time because of what they had learned from the previous situation.
Mr. Mehdi stated that the corporation had worked hard to incorporate safeguards and that technology had significantly advanced an interview from last week.
He remarked, “If something is incorrect, then you need to address it,” and then added, “We think we’re at the appropriate time to come to market and collect feedback.”
