Sam Altman: Openai was the “Bad Side of History” on the Open Source Code ncvrs.com

Opena researchers, engineers and executives, including Openai CEO Sam Altman, answered the questions to reduce product emissions Widespread Reddit Ama Friday.

Openai finds himself in a slightly uncertain situation. He fights with the perception of handing over to Chinese companies such as Deepseek, which Openai has allegedly stolen to his IP. Chatgpt manufacturer tried Sold up your relationship with Washington And at the same time, he is conducting an ambitious data center project, while allegedly creating a fundamental to one of the largest financing rounds in history.

Altman admitted that Deepseek had reduced Openai’s leadership in AI and said he thought that Openai was “the wrong side of history” when he opened his technologies. While Openai has open source models in the past, the company generally preferred a patented, closed source development approach.

“(Personally, I think we need to figure another open source strategy,” Altman said. “Not everyone shares this view, and this is not our current most important priority … we will produce better models (we move on) but We maintain less driving than in previous years. “

Kevin Weil, head of product at Openai, said in a follow -up response that Openai is considering open -purchased older models, which are no longer the most modern. “We definitely think of doing more,” he said, without being immersed in more detail.

In addition to making Openai to examine his expenditure philosophy, Altman said that Deepseek made the company to potentially find out the so-called O3-Mini, such as today’s O3-MINI Model, they show their “thinking process”. Currently, the Openai models are hiding their argument, which aims to prevent competitors from scraping their training data on their own models. In contrast, the Deepseek argument model, R1, shows its entire chain of thought.

“We are working to present a lot more than we show today – (showing the model’s thinking process) will be very quickly,” Weil added. “TBD everyone – showing all thought chains leads to competitive distillation, but we know that people (at least energy users) want it to find the right way to balance the balance.”

Altman and Weil tried to dispel the rumors that the chatgpt, the chatbot platform to launch many of the Openai models, will increase in the future. Altman said he wanted Chatgpt “cheaper” over time if possible.

Altman had previously said that Openai would lose money for the finest Chatgpt plan for Chatgpt Pro, which costs $ 200 per month.

In a slightly related thread, Weil said that Openai continues to see evidence that the more calculated performance leads to “better” and more performance models. This is largely what projects such as Stargate are needed, Openai’s recently announced huge data center project, Weil said. Serving the growing user base will also increase the need for calculation within Openai, he continued.

About recursive self -development that these huge models allow, Altman said he thinks that “fast take -off” is more likely than ever believed. Recursive self -development is a process in which the AI ​​system can improve its own intelligence and capabilities without human input.

Of course, it is worth noting that Altman is notorious for the overheating. Not long ago not he Lowered the Openai bar for AGIOr

One Reddit user asked whether the Openai models, self-development or not, are used for the development of nuclear weapons. This week, Openai announced a partnership with the US government that its models are partially handed over to the US national laboratories for nuclear protection research.

Weil said he trusted the government.

“I have met these scientists, and with world -class researchers, AI experts,” he said. “They understand the power and limitations of the models, and I do not think there is a chance that only a few model production will be done in a nuclear calculation. They are based on smart and evidence and do many attempts and data work to enforce all their work. ”

The Openai team was asked with several technical issues, for example, when Openai’s next argument model, O3 (“more than a few weeks, less than a few months,” said Altman); When the company’s next flagship, the “no justification” model, the GPT-5 can land (“no timeline yet,” Altman said); And when Openai can discover Dall-E 3 offspring, the company’s image-producing model. The Dall-E 3, which was issued about two years ago, was quite long in the tooth. Image generation technique has been upheavally improved since Dall-E 3 debuts and the model is No longer competitive for many reference testsOr

“Yes! We are working on it,” Weil said about tracking Dall-E 3. “And I think it’s worth the wait.”