Google’s AI-powered version generates completely implausible answers

Google’s AI-powered version generates completely implausible answers
Google’s AI-powered version generates completely implausible answers

New controversy for the Mountain View firm. The AI ​​Overviews feature, which provides artificial intelligence (AI)-powered answers in the Google search engine, has been generating numerous errors since its recent rollout.

Worrying false information

In just a few days, a flood of publications denouncing false, even grotesque, search results from AI Overview flooded the web. Announced with great fanfare at Google I/O, the tool generates summaries of information above traditional search results, saving the Internet user from having to navigate through several links to get the answer they want. .

For the moment, however, it is clear that it remains very imprecise. For example, he recommended inserting glue into a pizza recipe, or ingesting rock to extract nutrients. Untruths that are all the more worrying because they directly affect health. The AI ​​even advised to “ smoking two to three cigarettes daily during pregnancy “.

Most of the examples we saw were for uncommon queries, and we also saw examples that were doctored or that we couldn’t reproduce », defends Google in the lines of The Verge. The company maintains that its functionality delivers “ very high quality information “, but is forced to manually deactivate the tool for specific searches, as compromising screenshots are published on social networks.

Image: Google / YouTube.

Google is going through a tumultuous time

This new controversy highlights the difficulties encountered by the Mountain View firm in staying up to date in the race for generative AI. At the end of 2022, it was taken by surprise by the launch of ChatGPT then, a few months later, by the integration of AI into Bing.

The company then accelerated to compete with these two groups, especially since its main sector of activity, search engines, was directly threatened. The shift of younger users to TikTok or Instagram for search has increased the pressure on Google’s shoulders.

Bard’s rapid launch was heavily criticized internally and mocked online. The more recent release of Gemini also caused an uproar because the tool generated historically erroneous images. He has since been suspended. The AI ​​Overviews mishap only makes the situation worse, while Google is recognized for its technological advance and cutting-edge products.

The web giant cannot fail, as Google Search is used by billions of people to search for accurate, precise and quality information. The situation also serves as a reminder that generative AI is still far from being entirely reliable, with its suppliers still struggling to combat their propensity to hallucinate.



PREV The iPhone is controllable with the eyes in iOS 18
NEXT Good deal – The Netatmo connected object Connected thermostatic heads for radiators Additional “5-star” valve at €59.99 (-22%)