Die Rede auf dem geheimen Abendessen von Ultraman ist preisgegeben! Der CEO von OpenAI könnte möglicherweise ein KI-System sein. Ich möchte auch ein Chrome kaufen.

[New Intelligence Yuan Introduction] Altman outlined a grand vision at a dinner - from disrupting search and social media, to investing trillions in building data centers and new AI hardware, and even exploring brain-computer interfaces. He emphasized that AI is at a critical moment similar to the Internet bubble, but its potential is unparalleled.

Have you noticed that GPT-5 seems to have changed recently?

Just now, OpenAI announced that GPT-5 will soon restore the "warm" style of GPT-4o.

Altman believes that warming up the model is just the first step. Allowing users to customize the ChatGPT style can truly solve the problem.

But don't think the story of GPT-5 ends here.

At a dinner this Thursday, Altman revealed the astonishing blueprint of the "post-GPT-5 era." His ambition is simply off the charts:

From hardware, brain-computer interfaces, to disrupting social media, he even publicly said, "If Google really dares to sell Chrome, we'll have to take a good look!"

The demand for "human-created content" will rise significantly. The value of content created, recognized, and curated by humans will all increase sharply.

The most important milestone might be when most of our research clusters are occupied by AI researchers. This will happen gradually, and people will get more and more help from AI.

If we didn't have to pay for model training, we would have been profitable long ago.

I stopped using Google a long time ago. I can't even remember when I last used Google Search.

I'm not naturally suited to be the CEO of a listed company. Can you imagine me attending an earnings conference call?

Maybe in three years, the CEO will be an AI. Three years is a long time.

And, regarding the AI device jointly developed with Jony Ive:

It will take us some time, but I believe you'll find that all the waiting is worth it. I think it's simply incredible.

You know, new computing paradigms are extremely rare. There have probably only been two in the past half - century. So, get ready for a surprise. It's really worth the wait.

It can be said that the information revealed at this mysterious dinner is much more explosive than the model warming itself.

Altman's Dinner and His Trillion - Dollar Ambition

Next, let's follow the invited journalists to take a look.

On Thursday evening, at a Mediterranean restaurant in San Francisco.

We're sitting at the table, with Alcatraz Island outside the window. The fish entrees on the menu cost hundreds of dollars.

Just as we were chatting with other journalists, Altman walked in quickly through the door on our left.

He was looking down at his "naked" iPhone, trying to show us something. An inappropriate thought popped into our minds: "Going without a phone case, that's bold."

Of course, we soon realized that for this billionaire CEO who has recruited Apple's design veteran Jony Ive, preserving the original design of the iPhone is obviously more important than the $1000 cost of replacing the phone.

"Listen, we're about to launch an extremely beautiful device," Altman was referring to the AI device jointly developed by OpenAI and Ive. "If you dare to put a case on it, I'll come after you myself," he joked.

Altman invited about a dozen technology journalists to a "reportable dinner" and an "unreportable dessert" with him and other company executives.

However, this night left us with more questions than answers.

Unlike GPT-4, which emerged suddenly, far outpaced its competitors, and subverted people's imagination, GPT-5 performs at roughly the same level as the models of Google and Anthropic.

After many users expressed concerns about the tone of GPT-5 and its model routing mechanism, OpenAI even re - launched GPT-4o and the model selector.

But throughout the night, we became increasingly aware that the focus of this dinner was OpenAI's future beyond GPT-5.

The executives gave the impression that the importance of model releases is far less than when GPT-4 was launched in 2023.

After all, OpenAI is now a very different company. Its goal is to disrupt the traditional giants in the fields of search, consumer hardware, and enterprise software.

Altman mentioned that the proportion of users who have established an "unhealthy relationship" with ChatGPT is "far less than 1%".

But considering the current global users of ChatGPT, this number is definitely not small.

Altman admitted that OpenAI has held many internal meetings on this topic.

Some users really feel that they have established a relationship with ChatGPT. We noticed and started thinking about this problem a long time ago.

In addition, hundreds of millions of users, although they haven't formed a parasocial relationship, have gotten used to ChatGPT responding to them in a certain way, giving them a certain amount of emotional recognition and support.

Meanwhile, Altman didn't miss any opportunity to take a jab at Musk. He said:

You'll definitely see some companies making Japanese anime - style girlfriend robots because they think they've found a business opportunity.

But we won't do that. We'll continue to work hard to build truly useful applications and let users use them freely. But we'll also be careful not to accidentally take advantage of people with fragile mental states.

Altman hopes that ChatGPT can be as human - like as possible, but he doesn't intend to cater to specific ideologies or political stances.

We don't think our products should show a "woke" tendency, nor do we think we should go in the opposite direction.

We think our products should remain relatively neutral. If you want it to be very "woke", it should be able to do so. If you want it to be more conservative, it should also reflect that demand.

The number of ChatGPT users has nearly quadrupled in a year. Currently, it has over 700 million weekly users.

"We're now the fifth - largest website in the world. We think we'll soon rise to the third place (that is, surpass Instagram and Facebook). After that, it'll be difficult. It's very hard to surpass Google."

"Soon, billions of people will have conversations with ChatGPT every day. If the current growth rate continues, one day, the number of words spoken by ChatGPT every day will exceed the total number of words spoken by all humans."

The top 10 global websites statistics by SimilarWeb

To support the continuous expansion of its operation scale, OpenAI is in urgent need of more GPU resources, which is also one of Altman's top priorities at present.

We have to make very painful trade - offs now. We have better models, but we can't offer them due to insufficient computing power. We also have other new products and services that we can't launch either.

Altman also believes that we're in an "AI bubble" period.

When a bubble occurs, smart people get overly excited because of certain truths.

In historical bubbles, such as the tech bubble, there were also real things. The technology was indeed important, and the Internet was truly significant, but people still got too excited.

Are our current investors as a whole overly excited about AI? We think so.

But is AI one of the most important things in a long time? We also think so.

He confirmed recent reports that OpenAI is planning to fund a brain - computer interface startup to compete with Musk's Neuralink.

"We think neural interfaces are a cool idea worth exploring. We hope to be able to communicate with ChatGPT just by imagining."

Does Fidji Simo joining OpenAI to lead application development mean there will be other independent applications besides ChatGPT?

"Yes, you should look forward to it."

He also hinted at his social media ambitions:

We're very interested in whether it's possible to create a cooler social experience with AI.

Mastering the Future, Investing Trillions to Build an AI Empire

Altman has listened to the concerns from the outside world and learned from them, but he's pushing forward his plans with even more momentum - investing trillions of dollars.

After the release of GPT-5, Altman "listened to advice" and put the 4o series models back on the shelves.

Such a grand plan requires a huge amount of financial support.

Altman said, "It's foreseeable that in the near future, OpenAI will invest trillions of dollars in building data centers."

The following chart shows the growth trend of global data center capacity. By 2025, the power consumption of global data centers is expected to reach 485.4 terawatt - hours (TWh), accounting for about 1.7% of the global total power demand.

Altman admitted that the company may need to design a brand - new financing structure to raise such a large amount of investment.

"We think we can design a brand - new financial instrument for computing power financing that the world has never seen before," he said.

Regarding why he is investing so much in infrastructure, Altman's defense is: It's a good deal.

Our logic is that we can spend $300 billion and then sell $400 billion worth of services.

If we don't have these $300 billion data centers, we'll just keep disappointing our customers.

An important shift in the development of AI so far is that nowadays, more and more computing power is being used to respond to user queries (i.e., "inference").

"Most of the resources we're currently investing are used for inference," Altman explained. Inference refers to using computing power to run, rather than train AI models.

However, he also admitted that it's the cost of training new models that has prevented OpenAI from being profitable.

The inference business is profitable. If we don't count the training costs, we would be a very profitable company.

"We'll always train the next - generation models, but if our goal is just to be profitable and stay ahead, we think we can probably do it."

Altman compared the release of GPT-5 to the opening of Dickens' "A Tale of Two Cities": "It was the best of times, it was the worst of times."

"Some people will say, 'You took away our friend (GPT-4o). You're so mean.