Having released several new products on the eve of the Openai leaders, including the general director of Sam Altman, answered several questions on the Reddit platform. The head of the company said that she was ready to revise the attitude towards open source projects.
Image source: Dima Solomin / unsplash.com
The appearance of the Chinese Deepseek shared the position of Openai as a world leader in the field of artificial intelligence – it came to suspicions of theft of intellectual property. The ChatGPT developer seeks to strengthen his relationship with Washington and implement the Stargate project, which can lay the foundation for one of the largest rounds of financing in history. Altman admitted that Deepseek weakened Openai leadership in the field of AI and spoke out about open source projects – in this regard, the company, according to its head, was on the “wrong side of history.” Earlier, Openai issued open models of AI, but subsequently abandoned this.
«[Personally, I think] we need to develop another strategy regarding open source code. This point of view in Openai is not all shared by everyone, and now this is not our priority. [In the future] we will release the best models, but our leadership will become smaller than in previous years, ”Altman said. Openai is thinking of opening the code of already irrelevant models, added Kevin Weil Director for products. DeepSeek experience pushed Openai to make the mechanisms of the operation of current products more transparent: Deepseek R1 shows the entire chain of reasoning, and O3-Mini does not display this data. “We are working to show much more than today-[demonstration of the reasoning of the model] will appear very, very soon. So far, not all – the show of the entire chain will lead to distillation by competitors, but we know that people (at least experienced users) want this, so we will find a suitable way to balance it, ”Weil promised.
Image source: Growtika / unsplash.com
Altman and Weil tried to dispel the rumors that working with ChatGPT, a platform on which the company launches many of its models will soon rise in price – on the contrary, Openai would like to make the service even cheaper if such an opportunity is presented. Earlier, Altman said that the most expensive ChatGPT Pro subscription for $ 200 per month is for Openai losses. In a conversation with Reddit users, the company’s leaders also noted that with an increase in computing power, the performance and ability of models are growing – these considerations prompted Openai to initiate the Stargate project. According to Altman, the mechanism of recursive self -improvement – a process – a process in which AI independently increases its level without human participation, is also presented.
To the question of whether self -improved or any other Openai models will be used to develop nuclear weapons, Vail noted that the company trusts authorities working in government institutions by experts in the field of AI and world -class researchers – they are unlikely to dare to include AI answers in Projects with an extensive experimental base. Representatives of the company also answered other questions that are exciting public: the new reasoning model Openai O3 will be released “more than a few weeks later, and less than a few months later”; The deadline for the output of the flagship “non-aligning” model GPT-5 has not yet been set; The Dall-E 3 Image generator has been released more than two years ago, and now the company is preparing its new version.