To crown a day of product versions, researchers, engineers and OPENAI leaders, including the CEO of Openai, Sam Altman, answered questions in a large AMA Reddit AMA on Friday.
Openai is in a precarious position. He fights against the perception which he gives in land in the race of AI to Chinese societies like Deepseek, which, according to Openai, could have stolen his intellectual property. The Chatgpt manufacturer has tried to consolidate his relationship with Washington and simultaneously pursue an ambitious data center project, while posting the basics of one of the largest financing rounds in history.
Altman admitted that Deepseek had reduced the example of Openai in AI, and he said he thought that Optai had been “on the wrong side of history” when it comes to opening the ‘Supply of its technologies. While Openai has open observation models in the past, the company has generally favored a closed and owner -closed development approach.
“(I personally think we have to) find a different open source strategy,” said Altman. “Everyone in Openai does not share this point of view, and it is not our current priority … We will produce better models (in the future), but we will maintain less lead than we have done in previous years. “
In a follow -up response, Kevin Weil, Openai Products director, said Optaai envisaged older models from the opening of Open which are no longer at the cutting edge of technology. “We will certainly think of doing more,” he said, without going in more detail.
Beyond the incentive to Openai to reconsider his philosophy of liberation, Altman said that Deepseek had prompted the company to potentially reveal more about the way in which its so-called models of reasoning, like the O3-minini model published Today, show their “reflection process”. Currently, OpenAi models hide their reasoning, a strategy intended to prevent competitors from scratching training data for their own models. On the other hand, the Deepseek reasoning model, R1, shows its complete chain of thought.
“We are working to show a pile more than what we show today – (showing the model’s thinking process) will be very soon,” added Weil. “TBD on all – showing the whole chain of thought leads to competitive distillation, but we also know people (at least power users), we want it, we will find the right way to balance it.”
Altman and Weil tried to dispel the rumors that Chatgpt, the Chatbot platform through which Openai launches many of its models, would increase the price in the future. Altman said he would like to make the Chatppt “cheaper” over time, if possible.
Altman previously said that Optai loses money on its most expensive Chatgpt plan, Chatgpt Pro, which costs $ 200 per month.
In a somewhat linked thread, Weil said that Optai continues to see evidence that more computing power leads to “better” and more efficient models. This is largely what requires projects such as Stargate, the Massive Data Center project, recently announced by Openai, said Weil. The service of a growing user base also supplies the request for calculation within Openai, he continued.
Asked about recursive self-improvement which could be activated by these powerful models, Altman said that he thought that “rapid take-off” is more plausible than it thought in the past. Recursive self-improvement is a process where an AI system could improve its own intelligence and capacities without human contribution.
Of course, it should be noted that Altman is known to promise too much. Not long ago, he lowered the Openai bar to act.
A Reddit user asked whether OPENAI models, improving or not, would be used to develop destructive weapons – in particular nuclear weapons. This week, Openai announced a partnership with the US government to give its models to American national laboratories in part to nuclear defense research.
Weil said he had trusted the government.
“I got to know these scientists and they are IA experts in addition to world class researchers,” he said. “They understand the power and the limits of the models, and I do not think that there is a chance that they simply go a certain outing of the model in a nuclear calculation. They are intelligent and based on evidence and they have a lot of experimentation and data to validate all their work. »»
The Openai team has been asked several questions about a more technical nature, such as when the next Openai reasoning model, O3, will be published (“more than a few weeks, less than a few months,” said Altman); When the next flagship “non-resort” model of the company, GPT-5, could land (“I don’t have a calendar yet,” said Altman); And when Openai could unveil a Dall-E 3 successor, the company’s image generator model. Dall-E 3, which was released about two years ago, became quite long in the tooth. Image generation technology has improved by step as a giant since the beginnings of Dall-E 3, and the model is no longer competitive on a number of reference tests.
“Yes! We work there,” said Weil about Dall-E 3 follow-up. “And I think it will be worth waiting.”
