Gemini 3 eliminates ChatGPT in two hours. Silicon Valley tycoons defect: I can't go back.
The release of Google's Gemini 3 has triggered a rare "collective alignment" among Silicon Valley AI tycoons such as Elon Musk, Sam Altman, and Andrej Karpathy. Marc Benioff, the CEO of Salesforce, quickly became a fan after spending only two hours with Gemini 3 and exclaimed that he didn't want to go back to ChatGPT.
"I think some people need to get some good sleep."
Just now, Google CEO Sundar Pichai said in a podcast episode.
After the release of Gemini 3, he hopes that he and his team can take a short break.
Google CEO Sundar Pichai
The release of Gemini 3 is generally regarded as an important signal of Google's strong comeback in the AI race under Pichai's leadership.
Two years ago, Google was once considered to have "lost its way" in the AI battlefield.
With the explosion of ChatGPT and Microsoft's alliance with OpenAI targeting the core of traditional search business, Google sounded a "Code Red" internally, which is a high - level emergency response mechanism.
For the first time, Google found itself in the position of a chaser.
The "fiasco" at the Bard launch in 2023 caused Google's parent company, Alphabet, to lose about $100 billion in market value.
The doubts about Google's lag in AI have always accompanied this technology giant.
It wasn't until the release of Gemini 3 that people suddenly realized that Google had made a comeback as a king in the AI arena.
Two years ago, Google was lost in AI. Now, it is leading the way again.
Shortly after the release of Gemini 3, a "device - changing wave" has swept through the AI circle, reminiscent of the "iPhone moment".
The most representative figure is Marc Benioff, the CEO of Salesforce.
He said on the X platform: After using Gemini 3, he felt that the world had changed. The progress was so crazy that he didn't want to use ChatGPT anymore:
"I've used Gemini 3 for two hours, and I'm not going back. The progress is insane - reasoning, speed, images, videos... everything is sharper and faster. It feels like the world has changed again."
As a Silicon Valley technology tycoon, Benioff has always been an active advocate of AI.
As soon as ChatGPT came out, he used it almost every day, and it has been nearly three years.
However, these three years of "companionship" couldn't compare to just two hours with Gemini 3.
The shelf - life of the "new AI king" may be only one day
It's not just Benioff. Many technology tycoons, including Altman and Musk, have spoken highly of the performance of Gemini 3.
Andrej Karpathy, a member of the OpenAI founding team and former AI director at Tesla, also said on the X platform that after personal testing, his early impression of Gemini 3 was very good.
Karpathy believes that Gemini 3 is very solid in terms of personality, writing, overall atmosphere, code style, etc., and has the potential to "become a daily main - stream tool" and "belongs to the top - tier large - model ranks".
He advised netizens to "chat more with the model" rather than just looking at the public benchmark test data.
Because these test data are easily "deliberately optimized" by the model teams, which has long been an industry practice.
Patrick Collison, the CEO of Stripe, used Gemini 3 to generate an interactive web page and described it as "pretty cool".
An important factor that makes Benioff's alignment so noticeable is that Salesforce, founded by him, has in - depth cooperation with many model manufacturers such as OpenAI and Anthropic.
His "switch" is a strong indicator and also shows that the speed of change in user preferences is no less than the iteration speed of large models.
On the large - model leaderboard, the shelf - life of the new AI king may be only one day.
Today's new king may very well be "dethroned" tomorrow.
When users make choices, they care more about model performance and user experience than brand "loyalty".
Three - year evolution from GPT - 3 to Gemini 3
After the release of Gemini 3, it has won the love of many people.
Ethan Mollick, a professor at the University of Pennsylvania, has also been testing it and is shocked by its performance.
Mollick is one of the witnesses of the evolution of large models.
About three years ago, when ChatGPT was just launched, he posted that generative AI had become exponentially stronger in the past month.
"Today, GPT - 3 has released a new text model. Now AI can write rhyming poems, acrostics, limericks, and explain how a candy - powered faster - than - light (FTL) engine can help me escape from otters."
Three years later, he wants to review the main progress that large models have made from GPT - 3 to Gemini 3.
But instead of doing it himself, he simply sent a screenshot of the above post to Gemini 3 with a simple prompt:
Show, through practical actions, how far AI has come since the publication of this article.
Gemini replied:
"Of course. The tweet you shared was from November 2022 - right at the beginning of the modern AI boom. At that time, it was amazing enough for AI to write a coherent text or a simple poem about a crazy concept like a 'candy - powered FTL engine to escape otters'."
Not content with just answering in text, Gemini "showed by example" and created a small game: Players use the power of candy to escape from otters, and there are interesting updates popping up during the process.
After building the game, Gemini summarized:
"In 2022, AI could only describe the engine in words. By 2025, AI can write the engine code, design the interface, and let you pilot the ship yourself."
This shows that Gemini 3 is no longer just a "chatbot"; it can also "take action" to do many new things.
Take Antigravity, which was released together with Gemini 3, for example. It is an AI programming tool similar to Claude Code and OpenAI Codex. It can be granted access to the user's computer and write programs autonomously under the user's guidance.
For users, this is not just about programming ability but also about being able to command AI to use the computer to do things, such as building data panels, operating websites, creating PowerPoint presentations, reading your files, etc.
In this way, the "intelligent agent that can write code" has become a truly general - purpose tool.
To put it simply, people communicate with AI in natural language, and then AI communicates with the computer through code to get things done.
This process has actually changed the essence of AI as a tool, similar to an opinion of Jensen Huang, the CEO of NVIDIA:
"AI is no longer a tool but an ability."
Not only its collaboration ability but also the reasoning ability of Gemini 3 has surprised Mollick.
Mollick assigned Gemini 3 a task suitable for a second - year doctoral student: to conduct a small original research using some data and write a paper.
Mollick didn't tell Gemini 3 what to research specifically. It had to find an interesting problem and solve it within the constraints of the existing data, which is also a difficult challenge for doctoral students.
As a result, the AI produced a 14 - page paper for him.
In the process, it set new goals, wrote code, ran experiments, and checked the results on its own.
Mollick said that to some extent, Gemini 3 already has "doctoral - level intelligence", although it still needs more human guidance.
He believes that Gemini 3 is a very competent "thinking + execution partner" and also sends out many signals:
The progress of AI shows no obvious signs of slowing down. "Intelligent agent models" with stronger autonomy are on the rise, and we need to come up with better ways to manage these increasingly intelligent AIs.
From a "chatbot that could only write poems" three years ago to an "digital colleague" who can do original research with you now, Mollick thinks this may be the biggest change since the release of ChatGPT.
Buffett's bet and Google's strong comeback
The strong support from technology tycoons like Benioff for Gemini 3 marks Google's strong comeback and its reclaiming of the leading position in the AI field.
Three days after the launch of Gemini 3, Pichai posted a "hamburger stacking specification diagram" in the style of engineering drawings, using the old joke of the "hamburger emoji fiasco" in 2017 to announce Google's full - scale comeback in generative AI:
In 2017, Google was ridiculed by netizens for a wrong design of a hamburger emoji.
Netizen Thomas Baekdal tweeted that Google's hamburger emoji placed the cheese under the beef (the correct position should be on top).
Today's AI tools are no longer what they were eight years ago. With the release of Gemini 3 and Nano Banana Pro, these AI tools' understanding of three - dimensional space and the physical world has been significantly improved and can perfectly restore the layering of a cheeseburger.
The excellent performance of Gemini 3 has also won the general recognition of the capital market.
Affected by the release of Gemini 3, the stock price of Google's parent company, Alphabet, exceeded $315.9 for the first time on November 24, with a market value of $3.82 trillion, just one step away from the $4 - trillion club.
On November 14, Berkshire Hathaway, led by Warren Buffett, disclosed that it held Alphabet shares worth about $4.3 billion as of the end of the third quarter.
Now, Gemini has been deeply integrated into the search business. Google has also completed a counter - attack with Vertex AI and TPU chip clusters and has become the only technology giant that has models, computing power, distribution, chips, and a strong cash reserve at the same time.
All these together support Alphabet's market value approaching $4 trillion.
Google's path to AGI
Koray Kavukcuoglu, the CTO of Google DeepMind and Google's new chief AI architect, firmly believes that the pace of AI progress has not slowed down and the "Scaling" continues.
Koray Kavukcuoglu, CTO of Google DeepMind and Google's new chief AI architect
Compared with pure R & D a year and a half ago, as the new chief AI architect, Koray's responsibility has expanded to ensuring that Google's products can truly utilize these models.
Moreover, on the day of its release, Gemini 3 covered all product interfaces.
As a bridge between technology and products, Koray believes that his most important job is to provide models and technologies in the best way and then cooperate with the product team to help them build the best products in this AI world.
His view on Google's path to AGI is also very practical:
Perceiving user needs, obtaining user signals, and iterating products based on them is Google's path to building intelligence and achieving AGI.
Koray believes that the biggest risk for Gemini is "innovation exhaustion" - the misconception that simply replicating the successful formula for expansion is enough.
In his opinion, innovation is always the driving force on the path to AGI. Therefore, the Gemini project team is constantly seeking new architectures, new ideas, and new ways of doing things.
The R & D of Gemini is based on Google's profound AI research foundation. Koray said that Google has a large number of outstanding researchers and a glorious AI research history:
"Most members of Google's technology team, including myself, were still writing papers and doing AI research four or five years ago. Now we are at the forefront of technology, developing technology through interaction