Consider that your company has just hired a hot new executive, a rising star in the executive suite so seductive that a competitor has hired a lookalike. The energy around them is enthralling. From the CEO to the shareholders, everyone appears to agree that this person is the future of the company.
Then you find out the executive has a "hallucination problem" (Opens in a new tab). Every time they open their mouth, there's a 15 to 20% chance they'll make something up (Opens in a new tab). A Princeton professor refers to the man as a "bullshit generator" (Opens in a new tab). They literally can't tell the difference between fact and fiction (Opens in a new tab). They're heading on stage in five minutes to announce a new product. Do you continue to thrust them into the spotlight?
This week, the answer was yes for Microsoft and Google. Microsoft hosted a last-minute surprise event to reveal OpenAI would bring ChatGPT-style search to the Bing search engine and Edge browser, fueled by the success of OpenAI's ChatGPT, the Artificial Intelligence chatbot with 100 million monthly active users two months after its inception. Google launched its own AI search tool, Bard, the day before, and revealed it the next day at an event in Paris — but it ran into its own hallucinatory problem.
"A new race begins today," Microsoft CEO Satya Nadella told reporters gathered at the Redmond, Wash., headquarters on Tuesday. Isn't it lovely to think so (Opens in a new tab)? Microsoft, the chronically uncool kid on the computer block, would have you believe that Bing — sorry, "the New Bing" — is competing with Google in any search.
Google's pre-response announcing Bard(Opens in a new tab) was dripping with arrogance: "We refocused the company on AI six years ago," Google CEO Sundar Pichai wrote.
Google's 'hallucination problem'
That's a telling point. Google, the world leader in search, has had years to include AI, but its ChatGPT competitor, Bard, is still in beta with a small set of testers. Despite Pichai's hipster persona, the Bard reveal had an unintentional messiness to it. Google appears to have been taken off guard by the ChatGPT craze as well.
How else to explain the humiliating Bard gaffe on full show at launch — not at the event itself, where occasional demo hiccups are to be expected, but in a pre-made GIF? A user is depicted asking Bard for information about the James Webb Space Telescope that he may share with his 9-year-old son.
One of such "facts" is that the JWST captured the first image of an exoplanet. Bard was having hallucinations (Opens in a new tab). (UPDATE: While a Financial Times writer argues Bard's words were technically correct(Opens in a new tab), this requires a reading of the phrase that no human would ever apply – another another issue with AI search.)
It's no surprise that Alphabet's stock dropped as much as 8% on the day of the Bard launch. Google highlighted the key issue with AI search, implying that the corporation cannot use its massive stockpile of data to fact-check itself.
Google should know better, given that it already had a "hallucination problem" with highlighted snippets at the top of search results in 2017. The snippets algorithm seemed to enjoy telling lies about US leaders in particular. What could possibly go wrong?
ALSO READ: Never rely on a single source: The new ground principles for online learning
In other words, if you deploy an AI search tool too soon, you risk playing yourself. Microsoft was fortunate in that no glaring faults were visible at its debut ceremony. But, if ChatGPT-based search weren't fraught with flaws, why is it still in beta? Aside: There is a sign-up sheet if you want to do unpaid AI QA for Bing (Opens in a new tab).
"There's still more to do there," Sarah Bird, Microsoft's Head of Responsible AI (a revealing title!) responded to a Wired enquiry regarding ChatGPT's hallucination problem (Opens in a new tab). Yes, you read that correctly: the 15% hallucination figure came from a firm competing to construct a ChatGPT fact-checker (Opens in a new tab). (UPDATE: A breathless New York Times columnist's story on New Bing(Opens in a new tab) discovered that it couldn't even get basic math or a list of local kid-friendly activities straight.)
Previous versions of the program, according to Bird, might assist users in planning a school shooting, but this functionality had been removed. That's great to know! What more could possibly go wrong? There can't possibly be any more unexpected consequences hiding in this hallucinogenic experimental search tool that may disgrace a major and legally vulnerable digital corporation.
Clippy. Zune. Bing has been updated.
Of course, Microsoft understands embarrassment: it's the firm that brought us one of the worst blunders in software history, Clippy. The paperclip assistant was notorious for giving out unwelcome counsel. ChatGPT isn't like Clippy in the sense that we approach it with queries.
However, the fact that it frequently hallucinates its responses — or, more often than not, gives users a humdrum version on "I can't answer that" — might turn ChatGPT-enabled Bing into a type of Clippy on LSD. If enough casual users of "New Bing" get jumbled results, that's what they'll remember it for.
It makes no difference if a product improves later; the initial popular reaction is what might transform it into a punchline. Microsoft should know, because it gave us the Zune. It's the same as releasing a ChatGPT product before it's genuinely ready for primetime. https://ejtandemonium.com/
To be honest, "The New Bing" begs to be turned into a joke. Or are you really ready to abandon Google search and your Chrome browser in favor of Bing and Edge, should the latter win the AI search competition, whatever that means in this context? I didn't believe so. As a force, technological inertia is vastly underappreciated.
ChatGPT is impressive in some situations — real estate agents, in instance, use it(Opens in a new tab) for listing writing — but terrifying in others. But after you get past the headline, every report about its interruptions feels strangely diminished. It will result in a flood of student plagiarism! Except that it can also notify you when a paper has been authored by ChatGPT(Opens in a new tab), effectively negating its own threat. It passed a legal exam! Except that it barely passed with a C-plus (Opens in a new tab).
The problem is that creating the digital version of a human brain, called in AI circles as "general AI," is extremely difficult. We've only just begun to approach insect intelligence (Opens in a new tab), another long-held AI objective. Will you actually rely on ChatGPT to give your search results instead of, say, clicking on links yourself?
The solution may likely depend on how severe your hallucination problem is, dear reader. http://sentrateknikaprima.com/