Within 24 hours, after months of hype, Google and Microsoft announced the imminent arrival of Bard and a ChatGPT-integrated Bing search engine. At first glance, the public demonstrations by both tech giants appeared to feature potentially game-changing products that could revolutionize multiple industries. But it wasn’t long before even superficial reviews revealed egregious flaws in Google’s Bard proposals. Now it’s Microsoft’s turn to take a closer look, and the results are just as bad as Bard’s, if not worse.
Independent AI researcher Dmitri Brereton published a blog post Monday detailing numerous glaring issues in his experience with a ChatGPT-powered Bing. Bing’s demo often contained shoddy information: from inaccurate recommended product details, to omitting or misrepresenting travel stop details, to misrepresenting seemingly simple financial statements. In the latter case, Bing’s AI summation of basic financial data – something that Brereton says should be “trivial” for AI – contained completely wrong statistics out of thin air.
[Related: Just because an AI can hold a conversation does not make it smart.]
But even if it’s correct, Bing may have grossly skirted simple ethical guard rails. According to a report by PC world‘s Mark Hachman, the AI fed the Hachmans’ children a litany of ethnic slurs when asked for cultural nicknames. Although Bing preceded its examples by warning that certain nicknames were “neutral or positive, while others were derogatory or offensive,” the chatbot didn’t seem to bother to categorize its findings. Instead, it simply created a laundry list of good, bad, and extremely ugly offers.
This was announced by Microsoft’s communications director Caitlin Roulston The edge that the company “expects[ed] that the system can make mistakes during this preview period and the feedback is crucial to identify where things are not working well so we can learn and improve the models.”
As companies inevitably implement “smart” chatbot capabilities into their ecosystems, critics argue that it is vital that these issues are addressed and resolved before they become widespread. The missteps came as a complete surprise to Chinmay Hegde, an associate professor at the NYU Tandon School of Engineering, and Microsoft unveiled its technology far too soon.
[Related: Google’s own upcoming AI chatbot draws from the power of its search engine.]
“At a high level, the reason these errors occur is because the technology underlying ChatGPT a probabilistic [emphasis Hedge] large language model, so the output is fraught with uncertainty,” he writes in an email to PopSci. “We can never be absolutely sure what it will say next.” As such, programs like ChatGPT and Bard may be good for tasks where there isn’t a definite answer — such as joking or recipe ideas — but not so much when precision is required, such as historical facts or constructing logical arguments, says hedge.
“I’m shocked that the Bing team created this pre-recorded demo full of inaccurate information and confidently presented it to the world as if it were good,” Brereton writes in her blog post, before admonishing, “I’m even more shocked that the trick worked.” , and everyone jumped on the Bing AI hype bandwagon without doing an ounce of due diligence.”