This article is more than 1 year old

Microsoft to cap daily Bing AI queries to stop the bot delivering daft responses

Also: Google asks staff to rewrite Bard's responses; EU's AI bill may stall; Endangered architects

AI in brief Microsoft will start limiting the length of conversations with its AI-driven Bing chatbot to 50 turns per day, in a bid to prevent it generating unhinged responses to user queries.

People have reported the chat feature deployed on the company's Edge web browser going bonkers after being probed for too long. Bing has threatened users, gaslighted them, and become emotionally manipulative. It even claimed it was in love with a reporter – and told him to leave his wife.

(A good reporter will go to great lengths for a story, but you have to draw a line somewhere.)

In a bid to tame Bing's deranged behavior, Microsoft announced it was limiting users to 50 chat turns per day and no more than five chat turns per session. "A turn is a conversation exchange which contains both a user question and a reply from Bing," the company explained.

Although it's easy to think of Bing as having a personality, the chatbot is just software that doesn't understand what it's saying.

Trained to take on a conversational tone, it mimics human dialog and can start going off the rails if asked emotionally charged questions. Unfortunately, the AI can sometimes start doing this when asked seemingly benign questions too. 

"After a chat session hits five turns, people will be prompted to start a new topic. At the end of each chat session, context needs to be cleared so the model won't get confused," Microsoft clarified. 

Google asks employees to help train its AI search bot Bard

Google is asking staff to improve its AI chatbot, Bard, to be used for web search by rewriting text generated by the software. The ad behemoth's VP for search, Prabhakar Raghavan, reportedly sent an email to employees asking for their help in making sure Bard's responses are accurate and appropriate.

"This is exciting technology but still in its early days," he wrote, according to CNBC. "We feel a great responsibility to get it right, and your participation in the dogfood will help accelerate the model's training and test its load capacity (Not to mention, trying out Bard is actually quite fun!)."

"Bard learns best by example, so taking the time to rewrite a response thoughtfully will go a long way in helping us to improve the mode," he added.

CEO Sundar Pichai recommended employees spend about two to four hours of their time on Bard, and noted, "this will be a long journey for everyone, across the field." Although large language models like Bard are being hyped as a revolutionary way to search for information on the internet, they tend to generate unreliable responses and require clean data written by humans from which to learn.

Better data doesn't make these models perfect, but it often boosts their performance.

EU AI Act legislation vote could be delayed

The EU AI Act, aimed at regulating the technology to protect users from algorithmic harm and biases, is being pushed back as lawmakers are negotiating its final terms.

The bill was expected to be passed for member states to vote on at the end of next month, but it may be delayed since officials can't agree on basic principles, according to Reuters. The policies must be strict enough to counteract AI's lack of transparency and tendency to output biased results, while protecting data privacy – without suppressing development and investment. 

"The obvious tension here is between the focus on fundamental rights, on the one hand, and those who say these necessarily conflict with innovation," said MEP Sergey Lagodinsky, who represents the Greens party in Germany. Lawmakers are reportedly struggling to decide what AI applications should be labelled as "high risk", and subject to harsher rules. 

It's not clear when the bill will progress to the next legislative stages. "The file is long and complex, MEPs are working hard to reach an agreement on their mandate for negotiations. However there is no deadline or calendar on the next steps," one lawmaker said. 

Architects may be replaced by AI, warns professor

AI could take architects' jobs as image generation models get better at designing buildings, room interiors, and more.

Neil Leach, a professor at Florida International University who directs the Doctor of Design program, warned architects that commercial tools like DALL-E, Midjourney, and Stable Diffusion are putting their jobs at risk.

People can easily produce realistic renderings of houses, buildings, and rooms with text descriptions only, reducing the need to get architects to draw and model their designs. Leach said these tools do make architects more productive, but if workers can do more work in less time will companies still hire as many people?

"Often we are led to believe that rumors of the death of the architect are greatly exaggerated. The unique creative powers of the human mind, so the narrative goes, will endure. I beg to differ, however. There are signs that AI is becoming not only good, but terrifyingly good, to the point that it is beginning to expose our own limitations as human beings, and putting our jobs as architects unquestionably at risk," he wrote in an opinion piece for the design magazine Dezeen

Leach advised architects to pick up these AI tools and learn how to be a "superuser" to adapt as their profession evolves. ®

More about

TIP US OFF

Send us news


Other stories you might like