SearchGPT hallucinates answer in first demo video
“OpenAI’s SearchGPT Struggles in Its First Search Test” OpenAI’s new SearchGPT feature has hit a snag in its rollout, as highlighted by The Atlantic. This tool aims to enhance search capabilities by integrating OpenAI’s AI models with real-time web data to deliver faster and more accurate results. Although still in limited release, SearchGPT represents OpenAI’s vision for the future of AI-enhanced search.
During a prerecorded demonstration of SearchGPT, the tool provided incorrect information regarding the dates of the Appalachian Summer Festival in Boone, North Carolina. While SearchGPT indicated the festival took place from July 29 to August 16, the actual dates are from June 29 to July 27. Furthermore, as noted by The Atlantic and verified, the dates provided by SearchGPT correspond to the period when the festival’s box office is closed. You can view the complete demo here.
SearchGPT Still Looking
AI errors and “hallucinations” are common issues encountered across the board, a challenge that nearly every AI user faces. In this context, the recent mistake by OpenAI’s SearchGPT, while not unique, is noteworthy. The tool, designed to integrate real-time web data with AI models to provide accurate search results, falters in a way that somewhat diminishes OpenAI’s claim of reliability and transparency for SearchGPT.
This issue is not exclusive to OpenAI. For instance, Google’s AI assistant Bard (now known as Gemini) also faced criticism for inaccuracies. During a live demonstration, Bard incorrectly credited the James Webb Space Telescope with capturing the first images of a planet outside our solar system, while the actual images were taken by the European Very Large Telescope. This mistake contributed to a significant drop in Google’s stock price, reportedly costing the company around $100 billion.
In comparison, SearchGPT’s error is particularly striking because it occurred in a prerecorded video, which OpenAI had the opportunity to review and correct before release. This makes the oversight seem even more egregious than a live demonstration mistake.