OpenAI has launched a new product called SearchGPT, which it hopes will compete with Google.
It uses AI models similar to those used in the company’s more famous ChatGPT, but integrates them with real-time information intended to offer “fast and timely responses with clear and relevant sources.”
The company has stressed that the product is still in the initial testing phase and appears to have made a mistake in its announcement.
In the demo video, a user asked if there would be music festivals in a North Carolina city in August. The system gave an answer, but got the dates wrong, telling users that concerts would be held in August, when in fact that was when the box office would be closed. The Atlantic reported.
“This is an early prototype and we will continue to improve it,” OpenAI said. The Atlantic.
This blunder follows other high-profile mistakes made by similar AI systems. Last year, when Google launched the tool then known as Bard and since rebranded as Gemini, its demo video included a blunder about a NASA telescope taking the first image of a planet outside our solar system.
Both errors appear to be the result of the systems not fully understanding the material they were relying on. As with the dates presented by SearchGPT that were real but taken out of context, Google’s Gemini appears to have confused stories about NASA’s telescope that had been taken is First image of an exoplanet, this being the first photograph of one of them.
AI researchers have warned that such AI systems are prone to making mistakes because they struggle to understand context. They are also prone to “hallucinations” – giving answers that seem authoritative but are factually incorrect.
OpenAI said it was launching a preview of the SearchGPT feature to get feedback from a small group of users and publishers.
Google revolutionized its search engine in May with AI-generated written summaries that now frequently appear at the top of search results.
Summaries are intended to quickly answer a user’s search query so that they don’t necessarily have to click on a link and visit another website to find out more information.
Google’s makeover came after a year of testing with a small group of users, but it still produced errors that showed the risks of handing over information-finding to AI chatbots.
OpenAI’s close business partner Microsoft is also testing AI summarization in its Bing search engine.
Additional agency reports
