Google is manually removing the mistakes from AI Overviews

Google has admitted that it is “taking swift action” to remove the strange search results that its latest artificial intelligence tool has been generating. After the numerous examples of absurd answers posted by users on social media, the company has manually disabled AI Overviews for specific searches, while working to fix the situation.

Gemini DOWNLOAD

Launched in beta in May 2023 as part of the Generative Search Experience, AI Overviews has been tested for a whole year. According to Google CEO Sundar Pichai, the company has provided over one billion queries in that time and has managed to reduce the cost of delivering AI answers by 80% thanks to technical and hardware advances. However, this optimization seems to have happened before the technology was fully ready.

A founder of AI, who preferred to remain anonymous, told The Verge that “a company that was once known for being at the forefront and offering high-quality products is now known for offering low-quality products that are becoming memes.”

Despite this, Google insists that its AI Overview provides “high-quality information” overall. Meghann Farnsworth, spokesperson for the company, stated that many of the examples of incorrect answers were uncommon or manipulated queries, and that the company is using these incidents to improve its systems.

In addition, Gary Marcus, an AI expert and emeritus professor of Neural Sciences at the University of New York, highlighted the difficulty of perfecting this technology, stating that reaching 80% accuracy is relatively easy, but the final 20% is extremely difficult. Marcus and Yann LeCun, head of AI at Meta, agree that current language models will not achieve general artificial intelligence.

Gemini DOWNLOAD