OpenAI's entry into the AI search wars is long overdue. In the early morning of July 26, Beijing time, OpenAI opened the AI search function, and users can apply for internal testing on the SearchGPT official website from now on.
As early as the beginning of May this year, there were rumors about OpenAI's entry into the search engine market, but it was subsequently denied by OpenAI CEO Sam ·Altman. The official launch of the AI search function means that OpenAI is challenging Google's search supremacy head-on.
According to the official introduction, SearchGPT aims to enhance the dialogue ability of AI large models with real-time information from the network, providing users with fast and timely answers and clear and relevant sources. Users are able to ask questions about the initial answers given by AI searches. OpenAI said it plans to integrate the feature directly into ChatGPT products in the future.
SearchGPT演示截图
Judging from the above official profile, the general functional form of SearchGPT is similar to the existing AI search products on the market. As large models eagerly search for application scenarios, productivity tool scenarios, including search engines, have become the first stop on the ground. Internet giants, AI startups, etc., have entered the AI search track.
According to the statistics of the data platform Aicpb, in the Chinese domestic market, 360AI search, Mita AI search and Kunlun Wanwei's Tiangong AI search have the highest number of visits. In fact, chatbot products such as Kimi have similarly introduced the ability of AI search.
In foreign markets, Perplexity is attracting attention as a star company in AI search. The startup, which was founded in August 2022, has long ranked second in the AI search list in terms of monthly user visits, behind Microsoft's New Bing.
In late April, Perplexity closed a third round of funding of $62.7 million, valuing the company at more than $1 billion. SoftBank Group's Vision Fund II is investing in Perplexity at a valuation of $3 billion, with an investment of about $10 million to $20 million, as reported by Bloomberg in late June.
Behind the emergence of many AI search products, there are still many challenges such as the construction of commercial models, the accuracy of content generation, and the balance between the interests of online publishers.
Paid subscription or advertising models are the monetization methods in front of AI search manufacturers. Perplexity, YOU and other foreign leading AI search platforms rely on C-end payment for profit, and users need to open a membership to get a complete functional experience. However, Perplexity told the media in April that it planned to start selling ads.
Most of China's domestic AI search engine products are in the stage of driving traffic by free. However, a number of practitioners in the field told Nandu reporters that AI search may not be able to bypass the advertising model. A person from a domestic AI search manufacturer previously said that most domestic users subconsciously feel that search engines should be free, and the subscription route will be more difficult. "We have made internal judgments on the commercialization of AI search engines, and the best and smoothest is still the advertising model."
Although AI search can provide the source of the cited information, it cannot overcome the illusion problem of large models for the time being. Google, which launched AI Overviews in May this year, has frequently "overturned" in the accuracy of generated content. For example, when the "AI Overview" answers the question "cheese can't stick to the pizza", the user is advised to "pour some glue on the pizza"; mistaking former United States President Barack Obama for a "Muslim president", etc.
Liz Reid, Google's vice president and head ·of search, responded on May 30 that the "AI overview" not only provides text output, but also contains relevant links so that people can explore further. This means that the "AI Overview" usually doesn't hallucinate or make things up like other large model products. When the AI Overview goes wrong, it's usually because you misunderstand the question, misunderstand the nuances of language on the Internet, or don't have a lot of useful information available.
In response to the illusion of "AI overview," Google has chosen to scale back the application of AI search, such as "AI overview," which does not show hard news topics, "because the freshness and authenticity of these topics are important."
The conflict with the copyright interests of online publishers is also a problem that AI search manufacturers have to face.
In June, Forbes magazine publicly criticized Perplexity for plagiarizing Forbes articles at large lengths and for not quoting the original article prominently enough, which could lead some readers to mistakenly believe that the content was written by Perplexity. Perplexity, which came under pressure, later revealed that it had reached a revenue sharing agreement with premium content publishers.
On July 17, the Japan Press Association issued a statement saying that the AI search function launched by Google, Microsoft and other companies uses news content as an information source without permission, and many answers given by AI search are similar to the original news article, which is at risk of copyright infringement. And, many users don't visit recommended websites after they're satisfied with AI-generated answers, which creates a significant disadvantage for news organizations. The association requires AI search service providers to obtain permission from publishers and pay compensation, while calling on the Japan government to amend the law as soon as possible to improve the copyright law and other laws.
When OpenAI launched SearchGPT, it placed a strong emphasis on working with publishers and creators. OpenAI claims that SearchGPT helps users connect with publishers by prominently citing and linking original content in search sources, and users can quickly get more results from a sidebar with a link to the source.
SearchGPT侧边栏展示信源链接
"We work with publishers to create this experience and continue to seek their feedback." OpenAI has tried to show friendliness and has also introduced a way for publishers to manage how content is presented on SearchGPT, giving publishers more choice.
Chen Chen, a senior analyst at Analysys, previously told reporters from Nandu that the way AI search directly presents content may affect the interests of traditional content websites in the short term, but it will also give rise to new cooperation models and sharing mechanisms. For example, based on the capabilities of multi-modal large models, AI search service providers can develop advanced search subscription services and cooperate with content production agencies to provide customized content recommendations and in-depth services in professional fields.
Written by: Nandu trainee reporter Yang Liu