news

OpenAI's top management is in turmoil! Can the search business take a share from Google?

2024-08-07

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

Source: Visual China

Special author: Wu Qianyu

Editor: Ye Jinyan

Produced by: Deepnet·Tencent News Xiaoman Studio

August 6,OpenAICo-founder John Schulman tweeted that he would leave OpenAI to join another AI startup, Anthropic (OpenAI's competitor). He joined OpenAI at its inception nine years ago and was responsible for leadingChatGPTAnd API (application interface) post-training work.

According to foreign media reports, people familiar with the matter said that Peter Deng, vice president of OpenAI and head of ChatGPT, has also resigned. Greg Brockman, another co-founder and president of OpenAI, also said that he would take a long vacation until the end of this year.

OpenAI is now in a turbulent period, with internal turmoil, and The Information recently estimated that OpenAI's losses this year may be as high as $5 billion. Even so, OpenAI is still the hottest company in Silicon Valley, and talents are flocking to it. Under such circumstances, the business model is the sword of Damocles hanging over OpenAI's head.

On July 25, OpenAI officially released an AI search product called SearchGPT. Currently, the product is still in the testing phase. On the main page of OpenAI, there is an application entrance for the "waiting list". According to its spokesperson Kayla Wood, it will be open to 10,000 test users in the initial stage.

On the social platform, OpenAI CEO Sam Altman admitted that there is still room for improvement in today's search, and said that he likes this search method very much, which is more convenient than traditional search. As soon as the news came out, Google, the giant of "traditional search", was affected by the chain reaction, and its stock price fell by 3%.

From the PC Internet era to the mobile Internet era, after more than 20 years of development, the business model of traditional search engines has become very mature, and the dominant position of Google abroad and Baidu in China has also been basically stable.However, the emergence of generative AI is accelerating the "stirring up" of the search market with a solid interest structure.

On the day OpenAI launched its AI search, Microsoft Bing had already started internal testing of AI for a small number of users. Google had previously vowed to promote the AI ​​overview function in search to more than 1 billion users worldwide by the end of the year. Perplexity (valued at $3 billion), which was established less than two years ago, Baidu, 360 and other manufacturers are also betting heavily on AI search.

Faced with the fat piece of search, domestic and foreign big companies and entrepreneurs are eyeing it, and OpenAI's ambition has also been revealed. However, Google has long occupied the dominant position in the global search engine field, with a market share of more than 90%. Can OpenAI get a share of Google's pie?

In fact, OpenAI, which plans to become a for-profit company, is currently in a situation where it seems to have to grasp a "life-saving straw": GPT5, which was promised to be released in 2024, has not yet appeared, Sora has been relying on "futures" videos to "explode" in the circle, and a recent report shows that OpenAI may be on the verge of bankruptcy and its cash may run out by the end of this year.

On the other hand, the five domestic startups that emphasize big models with landing advantages (Dark Side of the Moon, Baichuan Intelligence, Zhipu AI, Minimax, Zero One Everything), the landing direction is not focused on search, but explored in a variety of directions. The reason, the CEO of one of the companies told AI Light Years, is that "there is no way to advertise."

OpenAI rushes into search

After SearchGPT was announced, a few users shared their experiences on social platforms.

According to Kesku's sharing, he gave some random prompts. For example, if you enter Nvidia stock, the Nvidia curve chart will be displayed, followed by a summary of the recent Nvidia stock market situation; if you enter Never Gonna Give You Up, the video and link on YouTube will appear, and there will also be a song introduction below.

During the experience, Kesku also found that local information can be provided through the user's IP or precise location, but this feature is turned off by default and can be selected in the settings - features such as "Movie theaters near me" are very useful.

In the comment section, someone asked, “How good is it compared to perplexxity?” Kesku demonstrated a case in which he asked SearchGPT and perplexxity respectively, “Who is Kesku?” The former could do it while the latter always failed, so he said that although he hadn’t done anything special yet, he currently liked SearchGPT in comparison.

From the demo released by test users and OpenAI, users can retrieve information in the same way as ChatGPT dialogues, and can further ask questions based on the answers provided by SearchGPT. But the difference is that SearchGPT can provide the latest information, and the relevant source links are also followed.

It is obvious that OpenAI attaches great importance to information sources this time. Not only does it show the reference link in the AI ​​summary when the mouse is pointed to in the demonstration video, but it also emphasizes the importance of credible sources and the need to protect information sources in the official blog post. Such caution may be affected by the frequent copyright disputes caused by ChatGPT. In addition, Perplexity was recently accused by publishers of directly plagiarizing their works due to an AI summary function, so OpenAI has to be more vigilant.

In terms of information presentation, SearchGPT is indeed very similar to perplexity, but SearchGPT displays more pictures and texts, and even videos. For example, in a demonstration, when a user searches for "Music Festivals in Boone, North Carolina in August 2024", SearchGPT will generate a list to display the results, with each music festival accompanied by its poster image.

OpenAI has always been in the spotlight. Whether it is a product update, personnel change or a demo, it will cause an "earthquake" in the industry. SearchGPT is no exception. Once released, it was dubbed as a "killer weapon" and "aimed at Google". However, when everyone studied the demo released by SearchGPT with a magnifying glass, OpenAI's self-proclaimed "authority" and "accuracy" did not stand the test.

A reporter from CNBN found that when answering questions about "music festivals held in Boone, North Carolina in August", SearchGPT gave the wrong ticket purchase date, and the answers given in succession were all music festivals in July and June that had already passed.

OpenAI is not the first to make mistakes in its demonstration videos. Previously, Sora’s videos were also pointed out to have many phenomena that violated the laws of reality, such as floating chairs, basketballs passing through the hoop, and inaccurate Japanese signs on the streets of Jingdong. However, as a startup, the losses of OpenAI’s “failure” are not so obvious.

Last year, when Google's AI big model Bard was demonstrated live, when answering questions about the new discoveries of the James Webb Space Telescope, it said that it had taken the first photo of an exoplanet. This was a serious inaccuracy, causing the stock price of Google's parent company Alphabet to plummet 9% that day, and its market value evaporated by more than $100 billion in two days.

Other testers also found problems with the accuracy of the content and pointed out that the Chinese comprehension ability was insufficient. Although OpenAI spokesperson Kayla Wood admitted the error and said that this was only an initial model and would continue to be improved,There are still voices in the market questioning whether OpenAI's entry into search was a bit hasty.

Betting on growth on SearchGPT

"A non-profit laboratory that is not controlled by anyone" was the original intention of OpenAI. This idealism was shattered in less than three years in the face of the high demand for computing power in reality. At that time, Musk wanted to merge OpenAI into Tesla, but the result was not satisfactory. In 2018, he withdrew from the team and stopped subsequent funding.

After that, OpenAI gradually developed in another direction, completely moving towards closed source and establishing a for-profit department... According to recent foreign media reports, Sam Altman is pushing the OpenAI board of directors to completely reform the company's non-profit structure to resolve the conflict between the company's non-profit nature and the multi-billion dollar revenue business it has established.

Currently, OpenAI's revenue mainly comes from three parts: ChatGPT subscription fees, API fees for developers, and To B cooperation income. In terms of commercialization, OpenAI has been taking action. Its previous cooperation with Apple, if it goes well, could reportedly bring in billions of dollars in revenue.

In order to increase revenue, OpenAI has also made other attempts, such as the previously highly anticipatedGPTs, and also benchmarks Apple stores.GPTThe dream of ChatGPT was realized. Before its official release, the traffic lost nearly 40%. In addition to GPTs, the traffic of ChatGPT has not fluctuated much since its explosive growth in early 2023.

The information recently estimated that OpenAI's losses this year could be as high as $5 billion. Combined with the $2.7 billion spent previously, the $10 billion Microsoft invested last year is no longer enough.

According to Microsoft's latest financial report, AI returns are lower than expected, and Intel's CEO also said that they have not yet fully benefited from the AI ​​trend. With heavy spending and a long payback period, OpenAI itself also needs to continue to reduce costs and increase revenue to maintain user growth. The launch of the search engine SearchGPT and the sudden launch of the voice function of GPT4.0, in the view of Liang Zhihui, head of 360AI search, is because GPT's traffic is not increasing, so it "cross-border robbery".

According to Business Insider, two people familiar with the matter said that under the leadership of Sam Altman, OpenAI may be able to launch GPT5 this summer. Now in August, OpenAI still has no signs of releasing GPT5. In an interview with AI Light Years, Zhu Xiaohu, a partner of Jinshajiang Venture Capital, doubted whether GPT5 had a significant technological leap in its core reasoning ability.

In addition, OpenAI's Sora has been sitting still as Runway, Kuaishou Keling, and pixverse have been intensively updating or releasing video generation products. In addition, last year, Sora's basic paper "Scalable diffusion models with transformers" was rejected by the top artificial intelligence academic conference CVPR due to "lack of innovation", and Sora's strength has also been questioned.

It now seems that OpenAI is betting its growth on SearchGPT. Will this product, which is far from perfect, change the market landscape of search?

OpenAI to the left, big model, five tigers to the right

Since ChatGPT exploded around the world, the technological competition between Internet giants and startups has intensified, and has gradually differentiated into two development paths: one is biased towards basic model capabilities, and the other is biased towards large-scale model applications. Since the beginning of this year, the latter has shown a strong ability to attract money, and many companies in AI search, video generation, and text creation have received financing.

The Dark Side of the Moon, Baichuan Intelligence, Minimax, Zhipu AI, and Zero One Everything, which are known as the "Five Tigers" of domestic large-scale models, have also received financing, and three of them are currently valued at more than 20 billion yuan. In terms of AI search, the "Five Tigers" are also making plans one after another.KimiIt launched a lightweight browser plug-in to provide users with intelligent search services; Baichuan Intelligence also released its first AI assistant "Bai Xiaoying", which focuses on "understanding search".

Wang Shaolan, president of Zhipu AI, said in an interview with AI Light Years and other media that AI+search is not a new topic and Zhipu has been working on it for a long time. "Whether it will become an innovation if we single it out today actually depends on future developments."

However, in the face of this trillion-level market, the "Five Tigers" did not focus on search, but explored in multiple directions, covering application categories such as chatbots, video generation, PPT, etc., involving many fields such as medical care, emotional companionship, office, media, etc.

In response to this question, one of the CEOs of the "Five Tigers" told AI Light Years bluntly, "There is no way to advertise." Search is close to money, which has long been an industry consensus. It is close because the business model is simple and relies on advertising for monetization. For example, Google's website advertising revenue will reach US$206.5 billion in 2023, Microsoft's search advertising revenue will reach US$12.21 billion in fiscal 2023, and Baidu's online marketing service revenue in 2023 will be 81.2 billion yuan.

However, the current AI search focuses on "no ads" to enhance user experience, which is tantamount to cutting off a source of income. As a star product of AI search, Perplexity launched a subscription payment model from the beginning, with a monthly fee of US$20 or annual fee of US$200. At the beginning of its establishment, it claimed to be "not affected by the advertising-driven model" and "avoiding complicated advertisements", but in April this year, it also changed its mind and said that it would launch an advertising business in the next few quarters.

Mita Search, known as the Chinese version of Perplexity, has the slogan "No ads, direct results." Previously, Fu Sheng, a shareholder of Mita Technology and chairman and CEO of Cheetah Mobile, mentioned in an interview with Geek Park that Mita AI Search may adopt a complex business model in the future and open up the in-depth analysis function to paying members. So far, there seems to be no clear commercialization path.

What’s even more cruel is that even as companies like Perplexity, Mita Search, and even Baidu, 360, and ByteDance have all entered the AI ​​search race, Google’s search market share remains high. According to Internet statistics agency Statcounter, as of June, Google held 91.1% of the search market share.

During conversations with many entrepreneurs, I found that they highly recognize the search track, but at the same time they are also in awe of AI search. On the one hand, they are concerned about how AI search can make money, and on the other hand, they are concerned about how to compete with giants.

Zhao Chong, founder of iDesign, which has received four rounds of financing, told AI Light Years that search is indeed a big market. Even if a startup gets 5%-10% of the pie, it is big enough, but market competition must be considered. "The giants are too big. A company with a valuation of more than 20 billion yuan will not have much chance to compete with them. But OpenAI's latest valuation has reached 80 billion US dollars, so it is possible to compete with the giants." As for commercialization, he said that forward payment, that is, subscription, and serving as a traffic entrance, introducing more ecosystems into it, and then monetizing through other products are all possible. "Essentially, it is advertising revenue."