SearchGPT, AI, research, implications, Researchers Create Self-Replicating Worm for GenAI Systems

Lately, when we search something on Google, a section titled “AI Overview” pops up at the top of the screen. This is Google’s AI-generated summary that provides quick, automated answers to queries. Behind the scenes, Alphabet has upleveled the Google search engine folding in GenAI capabilities to reduce work for the users. However, the feature faced a rocky reception as it started to surface hallucinated information and erroneous responses to some user inputs. Addressing the situation, Microsoft rolled back several features and retuned the algorithms at the backend. The AI Overview seems to be functioning much better since, but of course with some reservation. OpenAI’s SearchGPT seems to be on a parallel trajectory. Microsoft-backed OpenAI recently opened access to the temporary protype of SearchGPT, and users’ and publishers’ early impressions are optimistic but cautious.

The rising star of AI search, which is believed to be a direct challenger to Google Search, is designed to provide an interactive search experience.

“As a search engine built on generative AI, it will offer a brand new experience akin to the iPhone’s entrance into the mobile market,” said Jim Yu, founder and executive chairman of BrightEdge, a search marketing firm that published a report on SearchGPT analyzing its capabilities, and how the engine compares with Google AIO and Perplexity.

Powered by GPT-4 models, SearchGPT, much like ChatGPT, provides direct responses to queries, as opposed to pulling up a list of sources that contains the results.

Testers have said that the responses are well-rounded, concise and organized, and in many cases, outperform Google AIO’s results. Especially in areas like events planning and coding, tests reveal exceptionally refined responses.

AWS

SearchGPT marks sources of all information including links to webpages from where the information is derived. This provides additional depth and context for further exploration.

The tool’s ability to retain context allows users to make queries in a conversational flow as with ChatGPT, and keep digging deeper into contexts without launching new searches.

However, based on their experiences, testers have said that SearchGPT is still some ways away from being a threat to Google Search. The AI hallucination problem of ChatGPT still continues with SearchGPT.

Matt Berman, technology enthusiast and YouTuber, said there is a struggle with answering queries about online shopping and local events. In his review, he showed instances where SearchGPT turned out fabricated responses confidently.

BrightEdge’s study too mentions the problem of hallucination, but Yu estimates inaccurate responses to constitute only 1% or less of searches.

BrightEdge’s analysis focused on three metrics – the tool’s approach to queries, its ability to sense user intent, and its method of sourcing results.

On all three counts, it rated SearchGPT better than Google AIO and Perplexity, indicating that the answer engine is able to consistently interpret queries and surface comprehensive answers from authoritative sources.

“From unseen formats, such as visual responses, to a departure from the ad-based business model, SearchGPT is positioning itself as an answer engine that aims to improve the consumer experience. Brands who rely on Search to market and sell products and services online can’t afford to ignore the ‘searchquake’ – being present in the new AI engines is crucial,” Yu said.

Takeaways

With Google still holding over 90% of the market share, the barrier to entry for newcomers is markedly high. Besides Google’s portfolio of AI solutions in various niches – some of which were demoed at the last AI Field Day event –  provides a significant edge over its rivals.

As OpenAI eyes this Google-dominated search market that Microsoft has struggled to penetrate for decades, SearchGPT may offer a nice opening to get the foot in the door. The early feedback gives OpenAI the chance to address the shortcomings and tweak the models’ ability to understand and answer, making the tool reliable. However, taking a lead over Google is a distant possibility considering the behemoth’s sizeable investment and tenure.

TECHSTRONG TV

Click full-screen to enable volume control
Watch latest episodes and shows

AI Data Infrastructure Field Day

TECHSTRONG AI PODCAST

SHARE THIS STORY