Bard error wipes $100 billion off Alphabet’s value

The race to bring generative AI to the mainstream delivered a warning to Google today: an incorrect answer from Bard, its chatbot not yet generally available, has fueled doubts about the reliability of the demonstration the company was using to showcase it. Alphabet closed the session down 7.7%, a drop that wiped around $100 billion off its market value.

The error does not explain a market move of that magnitude on its own, but it arrived at the worst possible moment. Google is trying to convince investors and users that it can respond to the Microsoft–OpenAI alliance without sacrificing the reputation of its search engine, built over decades on its ability to find reliable information.

Bard attributed an achievement to James Webb that wasn’t its own

In promotional material released by Google this week, Bard was asked how to explain the James Webb Space Telescope’s new discoveries to a nine-year-old. Among its answers, the system claimed that Webb had taken the first images of a planet outside the Solar System.

The claim is false. The first direct images of an exoplanet were obtained in 2004 by the European Southern Observatory’s Very Large Telescope in Chile. James Webb, launched in December 2021 and scientifically operational since last summer, has opened up new ways to study the atmospheres of distant planets thanks to its infrared instruments. But it was not the first to photograph one.

This is no small detail for a demo designed specifically to show how Bard can answer educational questions and summarize scientific knowledge. Language models generate text from patterns learned across vast collections of data; they do not necessarily consult a verified database of facts each time they respond. That means they can produce a convincing explanation while introducing an incorrect detail. It is the problem known as hallucination: an invented or incorrect answer presented with an air of confidence.

Google races to catch up with Microsoft

Google introduced Bard on Monday and said it was opening the service to a group of trusted testers before making it available to the public in the coming weeks. The service is based on LaMDA, the company’s family of conversational language models, in a lighter version designed to serve more users.

Microsoft’s announcement, made on Tuesday, came a day after Google unveiled Bard. Microsoft has integrated conversational and writing capabilities into its search engine and Edge browser, turning a competition Google had dominated for years into a battle over the interface through which users access information.

The distinction matters: a conventional search engine provides links and lets readers cross-check sources; a conversational assistant tends to deliver a self-contained answer. If that answer is wrong, the error can easily go unnoticed, especially when the tone is fluent and seemingly authoritative.

The cost of failing in front of millions of users

Google knows this technology’s potential—and its risks—well. The company has been researching language models for years, but it has been more cautious than OpenAI about releasing them widely. Now it faces a different kind of commercial pressure: waiting too long could give Bing ground, but rushing a launch exposes flaws like Bard’s.

Alphabet’s decline reflects that tension more than any single incorrect fact. Investors are assessing whether generative AI will force a transformation of the search business, including the advertising that underpins much of Google’s revenue. Answering through a chatbot could change how much time people spend on a page, which links they visit and how ads are displayed.

Before Bard reaches the public, Google will have to prove two things at once: that it can compete on launch speed and that it has mechanisms to reduce errors in areas where an incorrect answer has real-world consequences. Today’s demonstration made clear that a more conversational interface does not remove the need to verify what a machine says.