OpenAI奇袭，谷歌反击

In the early morning of May 15, Beijing time, Google, which was "cut off" by OpenAI, an American artificial intelligence research company, held its annual Google I/O 2024 (2024 Google Developer Conference, hereinafter referred to as "Google I/O") as scheduled.

At the conference, Google released a number of new products, new tools, and new features, including: the latest large model Gemini 1.5 Pro has upgraded the context window from 1 million tokens (minimum input unit) to 2 million, and launched the lightweight model Gemini 1.5 Flash; Launched Imagen 3, a new model for Wensheng diagrams, Music AI Sandbox, and Veo, a video generation model. Incorporating AI into the search function, and launching "AI Overviews" in the United States; Project Astra, an AI general agent project, and Gemma2, Google's next-generation open-source model, were released.

If there is no OpenAI "cutting off", this will be the "boiling point" of this year's technology circle.

The day before Google I/O, in the early morning of May 14, Beijing time, OpenAI held a spring conference. The brief, which took less than 30 minutes, announced the new flagship model, GPT-4o, as well as more features for free in the larger model ChatGPT. However, GPT-4o's "human-like" response speed makes viewers feel that the implementation of VPA (virtual personal assistant) is just around the corner. This made the tech world boil ahead of time.

Wang Huainan, founder of Babytree and Micha Commune, stayed up late to watch OpenAI's spring press conference. He was also the CMO (Chief Marketing Officer) of Google Asia Pacific. He said that his old club (Google) held a large-scale product launch conference in California on May 14 (local time in the United States), "It must be AI-based." In his opinion, OpenAI used a seemingly casual 26 minutes to challenge a company's annual major release, "which is a four-to-two move".

Two hours later, Google hit back in a row

Compared with OpenAI's spring conference, which lasted less than 30 minutes, Google I/O spent nearly two hours announcing a number of new products and tools.

This year, Google I/O's main stage was set up in an amphitheater in California's Shoreline Lake Park. The day before the opening, Sundar Pichai, Google's chief executive, snapped a picture from the side of the stage and posted his first LinkedIn post with the caption, "Can't wait to see these seats filled with developers from all over the world."

Judging from the pictures related to the conference that have been leaked on the Internet, the amphitheater is full. At the opening of the conference, Sundar Pichai also said that more than 1.5 million developers are now using Google's native multimodal large model, Gemini.

But the race around AI is still ongoing.

At the conference, Google announced that Gemini, which was born for one year, has achieved a leap in capabilities, in addition to the 1.5 Pro advanced version of the context window expanded to 2 million tokens, the lightweight model 1.5 Flash has also reached 1 million tokens. Gemini also has a lot of expansions in terms of functions, such as mobile dialogue experience, conversation is more natural, the model can also be introduced into the bottom layer of the Android system, and the multi-modal Nano version of the model also includes fraud detection and other functions.

Google also launched three large model apps: Imagen 3, Music AI Sandbox, and Veo. Among them, Veo, a video generation model that can generate videos of more than one minute, is regarded as a product of Google's counterattack on OpenAI's Wensheng video model Sora.

谷歌还发布了AI通用智能体项目Project Astra和融进了Gemini的搜索功能AI Overviews。

Demis Hassabis, CEO of Google's DeepMind, made his debut at Google I/O. He presented Project Astra, Google's vision for the future of AI assistants, in which he filmed in real time while talking to an AI agent on his phone, demonstrating multimodal understanding and real-time conversational capabilities.

Judging by the conference video, Project Astra performed well, although it was a little slower than GPT-4o, which has a "human-like response time". If OpenAI hadn't been the first to release GPT-4o, ProjectAstra would most likely be the one that ignited the public's imagination of VPA.

Before OpenAI's spring conference, it was widely rumored that OpenAI would launch an AI search function. But Google, which is in the search industry, is the first to incorporate AI functions into its own search business.

Google said that based on the latest version of Gemini, users can ask the search engine anything they want to know or need to do, such as "finding the best yoga or pilates studios in Boston", in addition to giving search results, the search engine can also do a studio introduction, give information such as store distance and working hours; When users need to "create a diet plan", search engines can also do it.

"One of our biggest areas of investment and innovation is our founding product, search." Sundar Pichai reviewed the 25-year history of Google's creation of search, focusing on the improvement of Google's AI search achieved by Gemini's transformation.

After OpenAI was absent from the AI search track, Google made breakthroughs in multiple functions such as AI Overview, "Circle to Secarch" drawing and searching, and video search, so that its AI search, which supports multiple rounds of reasoning, planning capabilities, video questioning, etc., was displayed to the market and gave competitors a strong response.

Google's innovation crisis

Despite the unveiling of a number of updates and multiple products, it has not completely washed away the market's doubts about Google's innovative power.

Nearly two hours of Google I/O, Google's new products, new tools, new features, and even new infrastructure, etc., totaled more than 20 models, which also showed Google's emphasis on AI and its investment in the development of AGI (Artificial General Intelligence).

"The feedback on Google I/O has been very positive." A person who has worked at Google for more than a decade said.

But Mr. Wang said, "Google didn't catch the eyeballs. At the conference, Google, from the CEO to the general manager of many products, only three people appeared at OpenAI's spring conference to introduce GPT-4o, but in Wang Huainan's view, these three people "completely covered up the light of the two-hour Google I/O prepared by hundreds or even thousands of people behind them".

Wang Huainan said: "Today's Google is very much like the old Microsoft. "That's not a positive description. Because for people like Wang Huainan who came out of Silicon Valley more than 20 years ago, Microsoft "has no innovation, only knows how to make money, has no ideals, no mission, and chaotic products".

After reading Google's I/O, Wang Huainan said that although Google released a large number of new products, new tools, new features, and new infrastructure, the products released "are all products that protect their own business interests and protect search." At such a critical competitive juncture, almost a "revolutionary juncture", Google, which was once regarded as a benchmark for AI innovation, used a protective idea to protect its existing search habits and search business model, which reminded him of Microsoft back then.

But now Microsoft is moving briskly. Microsoft not only invested heavily in OpenAI's ChatGPT, but also tilted its Bing search servers, search data and even computing power to support OpenAI for large-scale model training, and a series of investments allowed Microsoft to "stride onto the AI revolution".

Wang Huainan said that he saw a Microsoft that "traveled lightly and did not rely on search to eat", and even found that the "old, slow, passively defensive, and fragmented" Microsoft had become flexible and user-oriented, just like the old Google that "amazed everyone from time to time with its innovation."

This makes Google's rival not only the young OpenAI, but also the changing Microsoft.

At present, Google, under the contrast of the bursting vitality of OpenAI and Microsoft, "urgently needs to break off its original thinking of innovation." But Wang Huainan also said that Google's current reinvestment in AI and organizational adjustment has allowed him to see the imagination.

Google, which started as a search engine, is now paying more and more attention to AI. At the conference, Google also counted that in the two-hour keynote speeches of Sundar Pichai and Damith Hasabis, AI was mentioned as many as 121 times, and the frequency of Gemini's appearance is not inferior to the word Google.

But Sundar Pichai also said at the conference that "Google is still in the early stages of AI platform transformation." Li Zhifei, the founder of Mobvoi, also said through social media that the current AI industry is also in the early stage, and whether it is technological development or business competition, "it is still far from the endgame".

Li Zhifei said that some of OpenAI's operations are "more and more opportunity-driven", in addition to scheming and Google to grab the headlines, the founder's state of "saying that AGI" is not optimistic about him. Based on this, he also said that on the road of running along the inertia and along the potential energy, OpenAI should also consider "how to avoid becoming a martyr in the AI era".