To create a day of product versions, researchers, engineers and executives, including CEO of Openai, SAM Altman, answered the questions in A. Wide -scale Reddit Ama Friday.
Openai finds the company itself in a risky place. She fights the perception that she waives the land of artificial intelligence for Chinese companies such as Deepseek, which claims Openai may have stolen its IP. Chatgpt maker was trying to do so The beach of its relationship with Washington Following up the project of an ambitious data center simultaneously, while it is said to have set the foundation for one of the largest financing rounds in history.
Altman admitted that Deepseek has reduced Openai’s progress in artificial intelligence, and he also said he believed that Openai was “on the wrong side of history” when it comes to its open techniques. While Openai has open source models in the past, the company has generally preferred an approach to the closed development of the source.
“(I personally think we need) discover a different open source strategy,” Altman said. “Everyone does not participate in Openai this opinion, nor is it not our current priority (…) We will produce better models (to go forward), but we will maintain less progress than we have done in previous years.”
In the follow -up response, Kevin Whale, chief product employee in Openai, said that Openai is considering the old open models that are no longer in the latest style. He said, “We will definitely think about doing more,” without going into more details.
Besides pushing Openai to reconsider the philosophy of its release, Altman said that Deepseek prompted the company to reveal more on how to show what is called thinking models, such as the O3-MINI model that was released today, the “thinking process”. Currently, Openai models hide their thinking, a strategy that aims to prevent competitors from stripping training data for their own models. On the other hand, the thinking model in Deepseek, R1 shows a series of full ideas.
“We are working on showing a group more than we are showing today – (showing the typical thinking process) will be very close,” he added. “TBD on everyone – showing all the series of thought leads to competitive distillation, but we also know that people (at least energy users) want it, so we will find the correct way to balance it.”
Try Altman and Weil to dispel rumors that ChatgPT, Chatbot app, is launched through which many of its bodies are launched, increasing in the price. Altman said he wants to make Chatgpt “cheaper” over time, if possible.
Altman previously said that Openai was losing money on the most batch Chatgpt plan, Chatgpt Pro, which cost $ 200 per month.
In a somewhat associated thread, Weil said that Openai continues to see evidence that more energy leads to “better” and more performance models. This is a large part of it that requires projects such as Stargate, the recently announced data center project in Openai. He continued that an increasing user base service stals the demand for account within Openai, as continued.
In response to a question about the self -improving self -improvement that may be enabled by these strong models, Altman said he believed that “rapid take off” is more logical than he believes. Self -improved self -improving is a process that can improve the artificial intelligence system intelligence and its capabilities without human inputs.
Of course, it should be noted that Altman is famous for excessive presentation. It was not long ago Reduce Openai’s tape for AGI.
A Reddit user asked whether or not Openai models, which will be used to develop destructive weapons-especially nuclear weapons. Openai announced this week a partnership with the United States government to give its models to the US national laboratories partially for nuclear defense research.
Will said he trusts in the United States government.
“I have met these scientists who are Amnesty International experts in addition to researchers worldwide,” he said. “They understand the strength and limits of models, and I do not think there is any opportunity because they are just some output in a nuclear account. They are smart and guided by evidence and do a lot of experiences and make data to verify the authenticity of all their work.”
The Openai team was asked several questions about a more technical nature, such as when the upcoming Openai model O3 (more than a few weeks is less than a few months ago, “said Altman), when the next pioneer of the company.” A non-expanded model, “GPT-5, (” You don’t have a timetable after “3, which was released about two years ago, has become old. The technique of generating images with jumps and borders has improved since the appearance of Dall-E3, and the model is It is no longer a competitive in a number of standard tests.
“Yes! We are working on that,” Will said about following Dall-E3. “I think it’s worth waiting.”
