Sam Altman: OpenAI has been on the ‘wrong side of history’ concerning open source

OpenAI CEO Sam Altman with OpenAI Researchers, Engineers and Executives answered a question to close one day of publication of the product in a day Broad Reddit Ama On Friday.

The OpenAI company has found itself in a slightly uncertain position. It is fighting the idea that it is in the AI ​​race DIPSECWhich Open has complained that his IP has been stolen. Chatzept creator is trying On the shore of its relationship with Washington And the same as followed by a Ambitious data center projectWhen the report has been reported Work on the basis For the largest finishing rounds in history.

Ultman acknowledges that DIPSEC has reduced the openings in AI, and he also said that Open is “wrong aspect of history” in the open-sourcing. OpenAAA has open-sourced models in the past, the company is usually a owned, in favor of the development of closed-source development.

“[I personally think we need to] Find out a different open source strategy, “said Ultman. “Not everyone in Openai shares this view and this is not even the highest priority of our current […] We’ll produce better models [going forward]But we will maintain less than the less leadership than the previous years. “

In response to a follow-up, Openai Chief Product Officer Kevin Wail said that Open-Sourcing is considering the old models that are no longer sophisticated. He did not go into more details and said, “We must think about doing more about it.”

Beyond requesting the Opeena to revisit its release philosophy, Ultman said that the DEPSEC has forced the company to express the company more possible on how its so-called rational models, such as published today and a 3-minute model, showed their “thought process”. Currently, Openai models hide their argument, a strategy is a strategy created by competitors to prevent training data for their own models. In contrast, the argument of the Dipsic shows the full thought chain of R1, R1.

“We are working to show more than we are today – [showing the model thought process] Will be soon, “Well added.” TBD is above all – all thoughts take the competitive patron, but we people (at least energy users) want it, so we will find the right way to balance it. “

Ultman and Well tried to remove the rumors that the Chattbott app, which launched many models of OpenAI, would increase prices. Ultman said that he would like to “cheap” Chatzipt over time if he was possible.

Ultman earlier said that Open is losing money at his principist Chatzipt Plan, Chatzipt Pro, which cost $ 200 per month.

In some related threads, Well said that Openi proof continues to see that the more calculated energy is “better” and leads to the more performance model. This is the need for projects like Stargate, Openai, in large part Recently announced Massive Data Center project, Well said. A growing user is increasing the demand for calculations in the opening of the base, as well as he also said.

Asked about the repetitive self-improvement that could be able to be able by these strong models, Ultman said that “fast -off” is more commendable than his belief. Repeated self-development is a process where an AI system can improve its own intellect and ability without human input.

Of course, it is noteworthy that Ultman is notorious for additional propaganda. He did not have too long ago The AGI has lowered the Opena’s barThe

A Reddit user asked whether the Openi models, whether self-development, would be used to develop destructive weapons-especially nuclear weapons. This week, the OpenAI has announced a partnership to give its model its model to US national laboratories in the nuclear defense research part with the US government.

Well said he believed in the US government.

“I got to know these scientists and they were world -class researchers as well as AI experts,” he said. “They understand the strength and limit of models and I don’t think they are some model output in nuclear calculations simply yolo yolo. They are smart and proven-based and they work a lot of experiments and data to legitimize all their work. “

The OpenAI team was asked several technical nature questions, such as the next logic model of Openai, and 3 (“more than a few weeks, less than a few months,” said Ultman), when the company’s next flagship “non-resoning” model, GPT. -5, can land (“there is not a timeline yet,” Ultman said, and when Openai can uncover the successor of Dal-E3, the company’s figure model. Has become The image generation has been developed by the leap and border after the launch of Dal-E3’s debut and the model There are no longer competitive in several benchmark examsThe

“Yes! We’re working with it, “Well said about a dal-e 3 follow-up.” And I think it would be like waiting. “