Top Stories

Google AI continues to have hallucinations. Is anybody concerned?


The experience of searching will be worsened by complacency about AI mistakes. Google ought to remove this latest function.


To "google" was, for years, to access the vast amounts of data on the internet. Wading through spam, advertisements, and, most lately, horribly erroneous AI responses is what it entails these days. Alphabet Inc. Chief Executive Officer Sundar Pichai must be wincing at the slew of blunders that have resulted from Google's new AI Overview tool, which has been live for the last week. In response to a question about cheese slipping off pizza, it suggested that the user use adhesive. One person took it to mean that a python was a mammal.


One of the most popular and lucrative technological products ever created has been severely compromised by Pichai, rendering it dangerously unstable. The timer is counting down until the moment he takes it down. It would be best if he did so sooner.


For the time being, Google has said that it is improving its AI search engine with each new hallucination report—which, according on my Twitter stream, seems to be becoming more and more frequent. Errors were occurring for "generally very uncommon queries and aren't representative of most people's experiences," a Google spokesman told The Verge. That's a weak justification from a business that takes great satisfaction in structuring the world's knowledge. Furthermore, because a large tail of rare searches make up the overwhelming majority of Google searches, infrequent search queries should provide trustworthy results.


This is a remarkable turnabout for a business that was previously so conservative that it wouldn't reveal its own generative AI technology, which was at least two years ahead of ChatGPT from OpenAI Inc. Since then, it has given in to the race that Microsoft Corp. and OpenAI started, which is igniting a never-ending dispute. A new version of ChatGPT was purposefully launched by OpenAI last week to coincide with Google's AI releases the following day. However, Sam Altman messed up the launch in the haste and had a falling out with Scarlett Johansson.


The 2011 tagline "It just works" by Steve Jobs summed up a time when dependability was the standard for technological goods. However, it will be increasingly difficult for tech firms to demonstrate the value of generative AI to both consumers and commercial clients the more examples they provide of how little it works.


Because his own SpaceX and Starlink companies are making blunders, even Elon Musk, who is on the cusp of raising $6 billion for his xAI venture, isn't employing generative AI technologies at these companies. He said at the Milken Institute conference earlier this month, "I'll ask it questions regarding Fermi Paradox, about rocket engine design, and about electrochemistry." "And the AI has performed horribly on all those questions thus far."


More false information will undoubtedly result if Google continues to support AI Overview and maintains the function. Another is that we'll become used to the absurd errors the AI makes, much as how we did when we scrolled past SEO spam and sponsored advertisements. There aren't many alternative choices, so we'll grow accustomed to an even worse service. (Google's search market share worldwide has decreased from 87% to 82% around ten years ago.) We now have to put up with shoddy software that was previously touted as a global revolution and needs ongoing fact-checking.


Although they are not a new issue, hallucinations seem to be one that we are unintentionally becoming used to. When errors surfaced during Google's first Bard demo in February 2023, Alphabet's stock fell 7%, wiping off $100 billion from the company's worth. They increased by about 1% on Friday as new social media pieces highlighting its most recent mistakes went viral. Wall Street seemed unconcerned. Does Google?


We'll find out when and if Pichai, as he did with the Gemini picture generator in February, stops his new AI function for more tweaking. He should simply do it to get tech back on the track of "just working," even if it would be another embarrassing setback.



No comments: