Google has become the “King of AI”. Is it all OpenAI’s fault?

In 2016, Google CEO Sundar Pichai proudly announced to the world at the I/O Developers Conference:

Google has become synonymous with search, and Alphabet will become an AI-first company in the future.

With its continuous efforts in deep learning, natural language processing and other fields, Google has indeed taken the lead at the starting line of this AI competition. However, the sudden emergence of ChatGPT has caused Google, which sounded the "red alert", to shift from peacetime to wartime preparation.

Many people optimistically predict that Google, the AI ​​mountain that has melted away overnight, will be able to regain its former majesty by next winter at the latest. But what was later staged was the story of a young man slaying a dragon.

OpenAI "steps on" Google to cross the river

If Google is crossing the river by feeling the stones of AI, then OpenAI is crossing the river by "stepping on" Google.

At the recent Google Next event, Google announced that the most powerful model, Gemini 1.5 Pro, is officially open to the public. It can support processing of up to 11 hours of audio and 1 hour of video, which made many onlookers excited. .

Some netizens uploaded a collection of the most iconic sports moments in history to Gemini 1.5 Pro, and asked it to identify the sports in the video, participating teams/athletes, years, etc., and it can clearly arrange it for you.

While people are still immersed in "playing" with Gemini 1.5 Pro, the competitive OpenAI announced the launch of the official version of GPT-4 Turbo. Not only has the basic capabilities been greatly updated, it also comes with a built-in image reading function. Now it has paid for all ChatGPT Open to users.

For a time, Gemini 1.5 Pro, which was supposed to be Google's killer weapon, became a "gone flower" again, giving people the feeling that Wang Feng joined Google.

Why do you say again? I can’t tell whether it’s intentional or unintentional, but this is not the first time OpenAI has sniped at Google’s new products.

In February this year, Google launched Gemini 1.5 Pro, the ultimate AI solution that it has been working on for a long time. It supports up to 1M contexts. In contrast, GPT-4 Turbo only supports 128K contexts.

The explosive headlines also lasted less than a few hours. OpenAI on the card table does not need to consider the rhythm of playing cards. Just take it out and you will find the king in the field of AI video generation – Sora.

Features such as minute-long video generation, real-time conversion of multiple lenses, excellent and stable picture generation, and world model understanding capabilities have attracted the attention of the world and suppressed the splash caused by Gemini 1.5 Pro.

Looking at the release pace of OpenAI products in the past, we can find that its powerful agenda-setting capabilities are undoubtedly the key words that cannot be avoided. Every major product launch has appropriately captured the market trends.

Previously, it was reported that while Anthropic was still weighing Claude's security issues, OpenAI had decisively relied on ChatGPT to seize the first-mover advantage in the market. After Anthropic received up to $4 billion in financing from Amazon, OpenAI once again focused on voice and image functions, giving Anthropic a slap in the face with ChatGPT, which can see, listen, and speak.

Behind all this may be inseparable from the influence of OpenAI founder Sam Altman. As early as when he was the head of YC, he showed a deep understanding and application of media influence.

According to the memories of The Information founder Jessica Lessin, Altman was particularly good at dealing with media reporters and used superb communication skills to win the trust of reporters. Even during OpenAI's court battle, he was able to control the direction of public opinion through a few tweets, and his influence was comparable to Musk in the Internet celebrity world.

Google is undoubtedly the unlucky one to fall victim to this Altman-style strategy. Why are these two headlines so important? This starts with the history of Google being “suppressed” by OpenAI.

Looking back on the past two years, under the aura of OpenAI, Google's AI journey seems a bit ill-fated.

At the end of 2022, ChatGPT lit up the entire AI world overnight, but Google, which got up early in the morning, was unable to do anything. Pichai looked at the "Manchu civil and military forces" in his AI field and wished he could stage a scene every day where Kangxi angrily denounced his ministers. Drama.

After letting ChatGPT, which had more than 100 million monthly active users, be "arrogant and domineering" for a long time, Google hastily launched a counterattack in February last year, but it still started badly. It confused the author of the first photo of an extrasolar planet by Bard, which made the audience who watched the press conference call out. The money also caused Google to instantly evaporate about 2.75 JD.com's market value.

If GPT-3.5 can still give Google a glimmer of hope to catch up, then GPT-4 has directly suppressed its peers for a whole year.

Thanks to the technical foundation accumulated over the years, the ship has turned around. After merging Google Brain and Google Deepmind, All In AI's Google has really created a "revamped" version of Bard's Gemini 1.0 model.

Among them, the ultra-large Gemini Ultra claims to completely crush GPT-4. However, just one day after its release, it was revealed that the demonstration video that broke the global Internet was fake, so much so that Google's own employees turned around and blew themselves up, saying, "This is not true." Realistic picture."

Until the emergence of Gemini 1.5 Pro, whether it was the technical ideas of the MoE architecture or the actual personal experience, in the praise of many netizens, it seemed to prove that the highest level of AI could be achieved by its ability to compete with GPT-4. Coming back.

In addition, the 1M context processing capability was even more powerful than its peers at the time. Gemini 1.5 Pro can instantly master an obscure language that only 200 people use, and can also understand a 44-minute Buster Keaton silent film in one sitting.

The two failures of "Toutiao" also made Google seem like the Sun Wukong causing trouble in heaven. No matter how much it jumped up and down, it could not get out of the palm of Buddha OpenAI.

It is worth mentioning that in terms of mainstream products, the two companies have also compared themselves relatively comprehensively. Regardless of the actual effectiveness, the main thing is that you have to have it and I have to have it.

  • You have ChatGPT, I have Bard (Gemini)
  • You have DALL·E 3, we have Imagen 2
  • You have Sora, I have Video Poet on PPT, and even Imagen 2 can create videos
  • Even in the field of search, Google's old business, OpenAI not only wants to replace it, but also has greater ambitions.

Not long ago, OpenAI CEO Sam Altman pointed out in an exclusive interview with Lex fridman that the world does not need another copy of Google.

The implication of his interview is that OpenAI will use LLM+ search to redefine the concept of search engines like Perplexity, and it will also find a way to make money that does not rely on advertising.

The person who knows you best is often your enemy

In the past year, OpenAI, which holds advanced AI tools such as ChatGPT and Sora, has become the strongest reference in the entire AI industry.

Just pick up a few large models from the large model manufacturers in the global "Thousand Model War", and terms such as "approaching GPT-4" and "comprehensive benchmarking against GPT-4" will make everyone laugh and make people laugh. The entire AI circle is filled with happy atmosphere.

Jim Fan recently said on social platform

Looking at the benchmark test lists and netizen reviews, the only ones that can gain a firm foothold and lead the trend in a short period of time are the four giants – OpenAI, Anthropic, Google and Meta.

The development of AI cannot bypass three major elements: computing power, algorithms, and data. Compared with Anthropic, Meta and Google, which have money, people and resources, are obviously more worth keeping an eye on. At the same time, compared to Meta, the relationship between Google and OpenAI is actually deeper than imagined.

The people who know you best are often not your friends, but your opponents.

Gu Long said this in the martial arts novel "Seven Weapons". This sentence also applies to the relationship between OpenAI and Google.

In the spring of 2017, an academic paper titled "Attention Is All You Need" accelerated the arrival of the generative AI wave. The content of the conversation between the author of the paper and Huang Renxun also became the focus of the entire AI community during GTC 2024.

What few people know is that all eight authors actually have the same identity—Google researchers. However, the advent of the Transformer architecture did not arouse enough attention from Google, as OpenAI CEO Sam Altman said:

When the Transformer paper was published, I don't think anyone at Google realized what it meant.

It was not until shortly after the paper was published that OpenAI became the biggest winner to get the first share of the pie, driven by the then OpenAI chief researcher Ilya Sutskever.

Today, the eight authors of the paper have all left Google. One of the core authors, Kaiser, has also joined OpenAI and is one of the inventors of a new technology called "Q*". This situation also happened to Sora, and the space-time patch technology behind it was also revealed to be standing on the shoulders of Google Deepmind.

In 2024, no technology company is willing to lag behind in the AI ​​​​track. The tug-of-war between OpenAI and Google, which has become more explosive, is a small profile of this trend.

According to The Information, Google co-founder Sergey Brin personally stepped down and made calls to retain employees who were about to switch to OpenAI. When OpenAI poached people from Google, it promised an annual salary of between US$5 million and US$10 million.

In addition, sources revealed that Hassabis, the head of Google Deepmind, complained that Google’s compensation policy made it difficult for him to prevent researchers from jumping to OpenAI.

Of course, Google is not passive. Logan Kilpatrick, the former head of developer relations at OpenAI, recently successfully joined Google. When Gemini 1.5 Pro was officially released, Kilpatrick also retweeted tweets to build momentum for his site.

From a financing background, the competition between OpenAI and Google is actually more like Microsoft behind the scenes playing against Google.

Looking at the mainstream large models on the foreign market, it seems that they cannot escape the fate of being absorbed by technology giants.

Apple has been buying a lot this year. Not to be outdone, Amazon chose to invest heavily in Anthropic, while Microsoft, which owns startups such as OpenAI, Mistral AI, and Inflection AI, is temporarily at the upper hand in the competition between these giants.

The feud between Microsoft and Google in the search era has also continued into today's competition in the AI ​​era.

In that era when search was king, Google firmly held the dominance of the search engine with its excellent algorithm and user experience. However, Microsoft's Bing has always been unable to shake Google's dominance.

According to data released by StatCounter, as of January 2023, Google's share of the global search engine market is as high as 92.9%, while Bing only has 3.03%. This huge difference in market share has allowed Bing to maintain its position as the "number two player in the millennium" for a long time.

Only this time, when the dawn of AI emerged, Microsoft, which had suffered setbacks in the search engine market, seemed to have found hope of making a comeback.

# Welcome to follow the official WeChat public account of Aifaner: Aifaner (WeChat ID: ifanr). More exciting content will be provided to you as soon as possible.

Ai Faner | Original link · View comments · Sina Weibo