Google vs. OpenAI

Round-the-clock technology

2024-05-17 17:06Posted on the official account of Shanghai All-weather Technology

In recent days, the biggest news in the tech world has been the two press conferences of OpenAI and Google.

Within 24 hours, OpenAI and Google, the two AI giants, showed their swords one after another.

On May 14, OpenAI CEO Sam Altman personally went down to create a momentum, and in a more than 20-minute press conference, he used the new multimodal model GPT-4o to awaken people's fantasies about the artificial intelligence system "Samantha" in the movie "Her".

The next day, Google announced that it would upgrade Gemini and launch the "AI Family Bucket", launch a search engine with integrated AI, and reshape Android with AI. It is trying to regain momentum and regain the initiative on the AI track. In the past year or so, OpenAI has repeatedly cut off Google, releasing new technologies and products before Google.

Whether it is the technical competition between the two sides or OpenAI's hand in hand with Apple, it shows that OpenAI, a popular star in the AI industry, has posed an unprecedented threat to Google.

Of course, Google's moat, which holds the search entrance and the ace of the Android ecosystem, is difficult to break in the short term. However, in the era of AI that subverts the industrial pattern, in the face of the backwave catching up, Google and other technology giants cannot relax, otherwise, an era is gone.

Warring

Despite being preempted by OpenAI, Google is clearly prepared.

OpenAI推出能说会道的GPT-4o，谷歌也有自己的萨曼莎——Project Astra。

"I've had this vision in my head for a long time." Demis Hassabis, head of Google's DeepMind and head of Google's AI efforts, said. Over the years, he has been thinking and researching artificial intelligence, thinking about how to make machines more human-like.

In his vision, an AI agent must be able to understand and respond to a complex and flexible world like a human. It needs to be able to see and respond to texts, and it must be active, educational, and individual, so that it can talk to it naturally and without delay. Eventually, all of this imagination became Project Astra. "It's that helper," said Hassabis, "it's useful, and you're used to it being there when you need it." ”

Judging by the demo video, it's clear that Project Astra is still far from what Hassabis envisioned. Even compared to GPT-4o, the latency is higher, and the communication is still not natural enough. But when Project Astra accurately said "glasses on the desk, next to the red apple", it still received an exclamation from the audience.

The emergence of Project Astra is also seen as a positive response to GPT-4o.

This isn't the first head-to-head battle between OpenAI and Google. In February of this year, a week after Google's press conference, OpenAI quietly dropped a bombshell and launched its first text-to-video model, Sora. Although Sora has not yet been opened to use, it still shocks the entire tech community.

For the bomb dropped by OpenAI, Google also countered at the developer conference and released the AI media creation model Veo. According to reports, Veo is capable of generating high-quality 1080p video and relies on Imagen 3, the latest text-to-image framework. These AI-generated videos can last more than a minute, and Veo is also able to understand specialized concepts in filmmaking and visual technology, such as time-lapse.

Not only that, but Google has integrated Gemini into almost all of its products, especially its strong business, search, to make search more powerful.

Although the two press conferences, one only lasted more than 20 minutes and the other lasted 2 hours, it was a technical pinnacle duel for the two companies.

anxiety

Behind the close hand-to-hand combat of the two AI giants is the commercialization anxiety of AI manufacturers.

In 2015, the little-known OpenAI was founded in Silicon Valley. At the time, it was a non-profit organization. Relying on Microsoft's $1 billion investment, he lived the life of a sweeping monk.

Since the advent of ChatGPT in November 2022, OpenAI has been an instant hit, and the commercialization process has been accelerating. In April 2024, OpenAI's COO said that the enterprise version of ChatGPT now has more than 600,000 registered users, compared to more than 150,000 in January this year.

Benefiting from the demand for AI technology from enterprises, it is reported that OpenAI's annual revenue in 2023 has exceeded $2 billion.

Under this AI fever wave caused by ChatGPT, major technology companies at home and abroad have sat down at the table.

On February 6, 2023, Google announced the launch of Bard, a conversational generative AI chatbot, the predecessor of Google's large model Gemini; Meta officially announced the release of Llama; In March of the same year, the American company Anthropic launched Claude; Domestic companies also immediately followed, Baidu released Wenxin Yiyan, Ali launched Tongyi Qianwen, Huawei's Pangu, and SenseTime is also new.

If OpenAI wants to go through the road of commercialization and seize the opportunity, it must ensure its technological leadership and continue to launch new and more powerful models. For now, however, it seems that this path has become more difficult.

After more than a year of development, various manufacturers have successively launched large models that benchmark GPT-4.

What's more challenging is that OpenAI's iteration is starting to slow down. From the first generation of ChatGPT to the release of ChatGPT, OpenAI only took more than 4 months. But more than 1 year has passed since the release of ChatGPT-4, and the much-anticipated ChatGPT-5 has not yet appeared. GPT-4o is impressive, but it's not enough to wow the industry as a whole.

Of course, OpenAI's slowdown is mainly subject to the objective law of large model iteration. The larger and stronger the model, the higher the data and computing power requirements. Eventually, this translates into cost pressures on OpenAI.

Compared with OpenAI's huge R&D costs, the annual revenue of more than $2 billion is still a drop in the bucket. According to Sam Altman, OpenAI is still losing money due to the high cost of building and running its models. As OpenAI develops more complex models, it is expected that spending will continue to outpace revenue growth. OpenAI may also need to raise tens of billions of dollars to meet these costs.

In addition to external competition, OpenAI also faces "internal worries". On May 15, Ilya Sutskever, who has not appeared since the "Gong Dou" incident at the end of last year, officially announced her departure, saying that she already has a plan for the next step, but cannot disclose details yet.

As the co-founder and chief scientist of OpenAI, Ilya Sutskever led the development of the GPT series of large models as well as the DALLE series of models. Although she was not present at the GPT-4o launch event, on the GPT-4o homepage, Ilya's name appears in the "Additional Leaders" section. The impact of Ilya Sutskever's departure on OpenAI's commercialization is still difficult to judge.

Remembrance

As the gap between technologies gradually narrows, the competition between AI vendors also enters the second half.

At the two press conferences, OpenAI and Google tacitly began to focus on the end side. GPT-4o launched a Mac version of the app, and Google reinvented Android with Gemini.

According to the analysis of China Galaxy Securities, the release of GPT-4o indicates that a new round of competition for AI Agent is about to begin, superimposed on the integration of ChatGPT on the macOS desktop version, and the end-side AI revolution is imminent, in addition, the B-end enterprise service market may usher in new growth opportunities because of the ease of use and inclusiveness of GPT-4o.

Under the new business opportunities, whoever can seize the opportunity and occupy more terminals will be able to occupy the market first.

In this regard, Google has a natural advantage. Google has a series of applications such as Google Search, Gmail, Youtube, Chrome browser, etc., and has built a strong ecosystem.

In contrast, as the first company to launch a large model, OpenAI has set a series of records under the AI dividend. Compared with Google, there is still a big gap between its user base and Google.

Similarweb statistics show that ChatGPT received 1.8 billion visits worldwide in April this year, up from March. Meanwhile, Google still saw 83.49 billion visits, even though it was down in April compared to March.

Not to mention, Google also owns Android. At this year's developer conference, Google upgraded Android Studio to integrate more AI features, making it easier for developers to call the Gemini API for software development.

Jim Fan, head of Nvidia's AI studio, said Google is doing one thing right and they're finally working hard to integrate AI into the search box. In his opinion, distribution is Google's moat. In this context, Gemini does not have to be the best model in the world to be the most used in the world.

In this war between OpenAI and Google, Google has a series of products such as Android, while OpenAI lacks an Apple.

As the first brother in mobile phone shipments in 2023, iPhone shipments reached 234.6 million units last year, with a market share of 20.1%. If OpenAI's large model can really replace "Siri" and become an AI intelligent assistant mounted on hundreds of millions of Apple devices, subverting the existing ecology in terms of search experience and efficiency improvement, for OpenAI, maybe everything will be different.