I ate dinner on Thursday Altman itselfSeveral other Openai directors and a petite group of reporters in San Francisco. Altman answered our questions for many hours. No topic was out of limits, and everything except what was said by dessert was on the album.
With a sporadic, broad interview with the Chief Director of Technology during a meal during a meal. But in a situation where Altman is, there is nothing common. Chatgpt quickly became one of the most used, influential products on Earth. Now Altman is planning an aggressive extension to consumer equipment, brain computer interfaces and social media. He is interested in buying Chrome if the US government forces Google to sell. Oh, and wants to collect trillions of dollars to build data centers.
But first he focused on answering the implementation of the GPT-5 last week. About an hour before starting dinner, Opelai pushed the update to restore “heat” 4o, the previous default ChatgPT model. It was Altman who called to quickly restore 4O as the option of paying subscribers after some protested with disappearance to Reddit and X.
“I think we have completely fucked up some things on the implementation,” he said. “On the other hand, our API has doubled within 48 hours and is growing. We are outside the GPU processor. ChatgPT has achieved a new level of users every day. Many users really love the switch model. I think we have learned lessons about improving the product for hundreds of millions of people in one day.”
He overcame the percentage of chatgpt users who have unhealthy relationships with the product in “Way below 1 percent”, but admitted that OpenAI employees have “many” meetings on this subject. “There are people who actually felt as if they had a relationship with chatgpt, and those people we were aware of and thought. And then there are hundreds of millions of other people who do not have chatgpt parasites, but were very accustomed to the fact that they responded to them in some way, and verify some things and would make some sense.”
“You will definitely see that some companies produce Japanese sexual bots of anime because they think they identified something here,” he said in a not very subtle dig. “You won’t see how we will do it. We will still work hard to create a useful application and try to let users use it as they want, but not so much that people who have really shortbread mental states have been accidentally used.”
Altman wants chatgpt to feel as personal as possible, but not necessarily play in a specific ideology or political view. “I don’t think our products should be awake. I don’t think they are something that is opposite.
Chatgpt approximately four times four times has pulled out the base of the user and now reaches over 700 million people every week. “Soon billions of people will talk to chatgpt a day,” said Altman. “We are now the fifth largest site in the world. I think we are on a clear road to the third.” (It means defeating Instagram and Facebook.) “Then it’s getting harder. For chatgpt to be bigger than Google, it’s really tough.”
For its operation to scale OPENAI, much more graphics processors need. This is one of Altman’s most crucial priorities. “You should expect Opeli to spend a trillion of dollars on the construction of a data center in a not very distant future,” he certainly said peace.
“We must now make these terrible compromises,” he said. “We have better models and we simply can’t offer them because we have no way. We have other up-to-date products and services that we would like to offer.”
He also thinks that we are in the AI bubble. “When there are bubbles, wise people are excessively excited about the nucleus of truth,” he explained. “If you look at most bubbles in history, just like a technological bubble, there was a real matter. The technology was really crucial. The Internet was a really substantial problem. People were excited. Are we in a phase in which investors as a whole are excessively excited about AI? My opinion is yes.
“You should expect Opeli to spend a trillion of dollars on the construction of a data center in a not very distant future”
He confirmed the recent reports that Opeli is planning to finance the startup of the cerebrospinal interface Elon Musk’s Neuralk. “I think that neural interfaces are cool ideas to investigate. I would like to be able to think something and make chatgpt answered.”
Does Simo Faids Attaching to Openai to launch the “application” means that in addition to ChatgPT there will be other independent applications? “Yes, you should expect it from us.” He suggested his ambitions in social media: “I am interested in whether it is possible to build a much cooler type of social experience with AI.” He also said: “If Chrome really intends to sell, we should look at it.”
Although Altman has a lot of interests, it is not really clear that operating OpenAI in the long run is one of them. “I am not a naturally well-matched general director of a public company,” he said at some point. “Can you imagine I’m on calling?”
Then I asked if he would be the general director in a few years. “I mean that maybe AI is in three years. It’s a lot of time.”
Here are some other things that Altman said:
- GPT-5 Creating: “We had this big GPU crunch. We could do another gigantic model. We could do it, and many people would like to use it, and we would disappoint them. So we said, let’s do a really smart, really useful model, but let’s try to optimize the costs of propellers. And I think that we have done great with it.”
- AI OPENAI device Z Ive ions: “It will take us a moment, but I think you will think that it is worth waiting. I think it’s amazing. You don’t get a new computer paradigm.
- The future of the network and publishers: “I think that people will go to fewer websites. I think that people will take care of the content developed by people than ever. My directional plant would be that the content of man created by man increased dramatically.”
- Which means agi: “Perhaps the most crucial milestone for us is that most of our research cluster is allocated to the AI researcher instead of human researchers. But I don’t think it will be so binary, because I think people will get a little more facilitate and a little more facilitate and a little more.”
- “If we didn’t pay for the training, we would be a very profitable company.”
- “I don’t apply Google anymore. I can’t say the law when I was the last time I was a Google search engine.”
Engaging career moves this week:
- I suppose I should start asking: “Are you going to quit your job?” in interviews. A week after mine Decoder The episode with him was published, CEO GitHUB Thomas Dohmke He announced that he was going on a startup life. The real story is that GitHub is less independent of the rest of Microsoft, which suggests that trade interests Jay Parikh’s The new AI team that absorbed him officially overtook Open Source Github, Switzerland for coding.
- Igor Babuschkinco -founder and de facto boss Elon Musk’s Business, He announced that he was leaving After two years of launching the investing company, it focused on AI safety. (I’m sure the last groc headlines had Thread Therefore with the time of this message.)
- Alexandr Wang in addition More OpenAI researchers in his new AI laboratory in Meta: Hyung won chungIN Jason WeiAND Zhiqing Sun.
- Added anthropically Dave orrGoogle veteran who recently conducted safety for Gemini as a security head. He also hired Jordan Burgess and his team during the startup of Humanloop, who worked on LLM ratings for companies.
- Joelle PineauFormer head of research AI META, he joined Cohere as the director of AI.
If you haven’t done it yet, don’t forget to subscribe The Vergewhich includes unlimited access to Command line And all our reports.
As always, I accept your opinion with satisfaction. You can answer here or Ping me safely on the signal.