If you have been following the debut of Microsoft’s ChatGPT-powered Bing chatbot, then you’ve seen just how much work still needs to be done. The Bing AI chatbot is impressive, but deranged, as it has been threatening, gaslighting, and outright lying to participants who have been accepted into the limited preview. Microsoft responded in a blog post on Wednesday, touching on some of its early observations and discussing improvements that need to be made.
Microsoft says it has seen more engagement across traditional search results following the launch of the new Bing. When users have received AI-powered answers, the results have generally been positive, with 71% of users giving the responses a “thumbs up.”
Although many users were satisfied with the new Bing, Microsoft also received plenty of feedback on areas that need work. It grouped those learnings into four categories.
First, Microsoft notes that Bing struggles when it comes to providing answers which require timely data, such as live sports scores or movie times. In order to improve these answers, Microsoft says it will send 4x the amount of grounding data to its model. The company is also considering adding a toggle to let you control the precision vs. creativity of answers.
Moving on to chat, Microsoft admits that it wasn’t prepared for so many people to engage with its chatbot as a form of “social entertainment” rather than a search tool.
“In this process, we have found that in long, extended chat sessions of 15 or more questions, Bing can become repetitive or be prompted/provoked to give responses that are not necessarily helpful or in line with our designed tone,” Microsoft explains.
Microsoft blames these AI issues on long chat sessions confusing the language model on what questions it’s answering and the model trying to respond in the tone in which it is being asked to provide responses. This is another problem that might be solved with “more fine-tuned controls,” such as letting the user “refresh” the context of the conversation.
Microsoft also highlights several bugs, such as slow loading, broken links, or incorrect formatting, many of which it says have been fixed in daily releases. Finally, the company acknowledges new features that users have requested, like booking flights, sending emails, and the ability to share searches and answers. Some of these will eventually make it into Bing.
