ChatGPT trapped in the webpage wants to break through in the mobile ecosystem

Source: Science and Technology News

Image source: Generated by Unbounded AI tool

"I believe that many of us have tried ChatGPT, and I believe that many people have put it aside after trying it, because it is basically separated from our work at present, so we put it down after using it. But I I still hope that everyone will not 'get up early and catch up on the late episode', because this is a paradigm revolution that will bring subversive changes."

Recently, Wei Qing, chief technology officer of Microsoft China, said the above passage at a new book reading session. The second half of the sentence still needs time to verify, but the "separation" of the first half of the sentence has become a common reality.

ChatGPT has been on the Apple App Store for a full month and has won 5 million downloads. But its web traffic entrance is gradually slowing down. The previous miracle of high-speed growth has become short-lived.

The total number of users is limited, and the faster the growth rate, the shorter the time to reach the ceiling. In less than a year, ChatGPT, which has already encountered a traffic ceiling on the PC side, is aiming at the mobile Internet in the next step.

Desktop ecology peaked

Even if ChatGPT soars like a rocket, it will eventually have to face the laws of gravity.

According to data from the analysis agency SimilarWeb, ChatGPT’s web page visits have begun to experience a slowdown in growth. Although the global visits still reached 1.8 billion in May, the month-on-month growth rate was only 2.8%.

Compared with the three-digit month-on-month surge since January, the concept of a large model with a force field that distorts reality in the past is returning to the physical world.

This is not only a problem that ChatGPT itself is facing, but Bing, which has the same large model core, is also returning to the old track.

The data insight from SimilarWeb shows that when Bing introduced the GPT-3.5/4 model in February, in the natural search keywords from March to May, users used keywords such as New bing, which represents the large model version, to search for the incoming natural traffic. A sharp drop of 56.84%.

On the contrary, the natural traffic keywords related to traditional search engines are returning, which also means to a certain extent that search engines empowered by large model capabilities are not so easy to use.

Taking market share as an example, the growth rate expectations of large models represented by New bing continue to fall short. In March this year, the market share was 2.86%, which dropped to 2.77% in May, returning to the normal market fluctuation range.

At the same time, competitors of the same type began to copy the initial growth rate of ChatGPT to catch up.

For example, Character.Ai, the second largest artificial intelligence chat website, had 281.4 million visits in May, a month-on-month increase of 62.5%; Google Bard, which was launched in February this year and fully opened in early May, had 142.6 million visits in the same month, a month-on-month increase of 187.2% %.

Although the two are still far behind ChatGPT’s 1.8 billion visits, ChatGPT is obviously facing more latecomers in the competitive ecology of PC desktops.

New opponents emerged one after another, and old partners also made troubles. ChatGPT, which "grabs food" with a subversive user experience, is an out-and-out barbarian in the eyes of traditional websites.

In May alone, according to incomplete statistics reported by the media, three companies were the first to bear the pressure due to the growth of ChatGPT’s traffic.

First, the traffic of Reddit, the largest social news forum in the United States, declined, and then announced a fee for the API. As OpenAI's earliest source of large-scale model training data, Reddit tries to make up for lost commercial benefits in this way.

Coincidentally, Stack Overflow, one of the most commonly used websites for programmers, also encountered the same dilemma.

SimilarWeb, a web analysis company, reported data in April that since March this year, Stack Overflow's traffic has dropped by 13.9% compared with February, and the site's traffic in April has dropped by 17.7% compared to March.

From long-term single-digit fluctuations in 2022 to double-digit declines for several consecutive months this year, the main reason is that more users use ChatGPT to help debug code. The programmer segment is currently the area with the highest usage and frequency of ChatGPT.

Chegg, a vertical academic forum website, was previously popular among college students. Because it helps students with complex homework assignments like those encountered in STEM classes. Like Stack Overflow, users post their homework questions and then wait for a human to respond.

However, this business logic is under pressure from ChatGPT, which can provide students with instant answers. Chegg recently reported on an earnings call that the company's share price and revenue are declining, with ChatGPT being the main reason.

Chegg CEO Dan Rosensweig said in a conference call earlier this month: “Since March, we have seen a significant increase in student interest in ChatGPT. We now believe it is having an impact on our new customer growth rate. "

Under the PC desktop ecology, in the face of the pressure of ChatGPT, Stack Overflow and Chegg have successively announced measures such as API payment and the development of large models in vertical fields to raise the threshold of desktop data circulation. This is also true of the industry as a whole.

Take the domestic market, which is the most active for large models, as an example. According to the "China Artificial Intelligence Large-scale Model Map Research Report" released by the China Institute of Scientific and Technological Information under the Ministry of Science and Technology in May, as of May 28, at least 79 domestic basic large-scale models with a parameter scale of 1 billion or more have been released.

However, according to Caijing, many people in the industry believe that there are very few companies that have the actual ability to run through the commercialization of large models. When the growth rate of ChatGPT, the industry leader, slows down, the large model is also coming to the end of the release of PC desktop products: breaking away from the freshness of the market in the early stage, and entering a new node of landing with a more competitive environment.

Some American analysts commented: "In the history of technological products, there has never been a lack of epoch-making technological novelties, such as the first year of VR, the first year of autonomous driving, and the first year of artificial intelligence that have repeatedly appeared. But most of the time, these The so-called first year is finally waiting for the next restart after the early adopters psychology that has swept through users around the world.”

For ChatGPT, after sweeping the global users with the technological experience of a large model, although it is not so pessimistic to wait for the restart, how to find more scenarios to retain users is imminent.

Beach Landing Mobile

OpenAI previously launched the ChatGPT desktop plug-in, but it has not achieved significant growth. Founder Sam Altman said in an interview that ChatGPT will not release a follow-up plug-in soon, because from the actual market situation, the plug-in has not yet reached the product-market fit.

Perhaps it is the setback of the desktop plug-in, which has made the mobile scene the main action of OpenAI in the past month.

It was launched on the Apple App Store on May 18. So far, it has continuously updated 6 minor versions. Compared with the web version of ChatGPT, the app version in the mobile ecosystem is obviously more diligent, and of course the benefits are obvious.

First of all, from the perspective of traffic ceiling. According to the global Internet traffic distribution survey released by ExplodingTopics data service provider in June, users from mobile devices currently contribute nearly 55% of global website traffic.

This means that the mobile ecology can quickly raise the traffic ceiling for ChatGPT's large-scale enclosure movement.

Second, there are more users to reach. About 92.3 percent of users use mobile phones to access the internet; only 65.6 percent use more traditional laptops or desktops, the data showed.

Finally, according to the division of regions, the development of mobile ecology is also conducive to ChatGPT's further sinking of large-scale model capabilities with the help of the mobile Internet with a higher penetration rate.

At present, ChatGPT's large-scale model users are mainly distributed in developed countries such as the United States and Japan. For most developing countries such as Africa and Asia, the penetration rate and usage of mobile devices are much higher than that of desktop devices.

According to overseas media estimates, after the release of the Android version, with the help of the high penetration rate of the mobile ecosystem, ChatGPT can theoretically cover 86.29% of the global population.

However, due to the more mature software and hardware closed-loop ecology of the mobile terminal, the current progress of ChatGPT's App version can hardly be called smooth.

According to public reports, on the day of launch, Apple announced restrictions on the use of ChatGPT and other external AI tools within the company. Apple is concerned that employees using external programs "may reveal confidential data," the document states.

Even so, grabbing the beach and moving the scene is still a key step in the large-scale enclosure movement. Various manufacturers are trying to release smaller versions of the big model to run on mobile phones.

Taking Google as an example, the new generation of large language model PaLM 2 released has four versions, from small to large, namely Gecko, Otter, Biso and Unicorn. Among them, Gecko can run on a mobile phone, and it is fast enough to work without an Internet connection.

Baidu in China released its first smartphone, which is positioned as a learning machine for teenagers, with Wenxinyiyan as the core, equipped with AI learning model, and claims to be a new species of smartphone.

Large model +, the smooth road has not yet appeared

From ChatGPT giving new life to the search engine, the growth rate slowed down after 6 months; from publishing the large-scale model of the web page together to grabbing the mobile scene application; under the wave sweeping the world, the speed of redoing everything on the large model is not as fast as imagined quick.

According to the technology media The Information, after exploring large models + search engines and large models + Apps, OpenAI is planning to launch a large model application store similar to the App Store.

The store allows developers to list their products based on OpenAI technology, such as chatbots or custom models for various vertical fields, providing enterprises with a way to access cutting-edge AI large models.

This path means that OpenAI will jump out of the previous empowerment path of existing Internet products optimized and improved with large model capabilities, and instead build a new AI ecosystem around developers.

Under the role of an ecological builder, not only a general multi-modal large model such as GPT, but also an open connection to the vertical domain model of GPT. It seems that instead of being a centralized traffic portal, OpenAI wants to be a platform.

Remarkably, nearly all of the most valuable companies in the world operate digital platforms that support billions of daily users and tens of billions of dollars in economic value every day.

iOS is owned by Apple, the world's most valuable company, while Android is owned by third-largest Google. Microsoft owns Windows, the third-largest operating system, and Azure, the second-largest provider of cloud infrastructure.

Meta, the ninth largest company in the world, owns the world's largest online social identity platform, with thousands of websites and apps using its "login with Facebook" feature in place of or alongside its proprietary customer ID system.

Obviously, as an epoch-making person who launched a large model, such ambitions are more in line with the potential of OpenAI.

However, being a leader in building a platform means that for more large-scale model practitioners, the road to the commercial application of "large-scale model +" has not yet appeared smooth. Issues surrounding the distribution of data and commercial interests will also emerge endlessly in the future.

Taking "big model + education" as an example, before Xueersi and its partners were exposed to "scraping the database" incident, the core contradiction was the data partner, who believed that Xueersi illegally used its data to develop the large model of AI composition assistant.

In the future, with the increasing scale, complexity and importance of AI technology represented by large models in the global digital economic system, the construction of AI business civilization and norms is also the top priority for the regulators above the market. Heavy.