Openai announced the launch of his new model, called O3-mini, a few days ago. That event was accompanied by a Ama Session (Ask Me Anything) in which several managers of the company, including its CEO, Sam Altman, answered the questions of Reddit’s readers. And among his statements, one especially striking.
Open Source. In one of the questions that Altman admitted that Depseek had reduced Openai leadership in the field of artificial intelligence, and also added that he believed that Openai had been “in the wrong place in history” when publishing his Open Source license technologies.
OpenAi has little Open. The truth is that although Openai did have much more consistent beginnings with his name. GPT-1 and GPT-2, its first models were published with much greater transparency and inviting the entire community to benefit from them. The thing changed with GPT-3 and successive models in which it barely shared technical details to prevent competitors from using them for their benefit.
But publishing an oun source is in the air. Altman admitted that internally he was being discussed to publish the weights with which his models train – an interesting information for the community. However, he also stressed that “not everyone in Openai shares that vision (opening)”, and pointed out that a potential change in strategy in this regard is not currently something of high priority. “
Maybe ancient models become Open Source. Kevin Weil, one of the top responsible for OpenAi’s models, indicated that the company is considering publishing as Open Source older models that are no longer up to the last. “Of course we are thinking of doing more about it,” he concluded without giving more details or estimated dates.
Deepseek has changed the panorama. The OpenAi CEO also highlighted how its company intends to develop better models little by little, worse admitted that its leadership in the AI scope will no longer be as clear as it had been so far, and the reason is Deepseek.
And he has forced O3-mini to teach how he thinks. In fact, the appearance of Deepseek R1 has caused Openai to decide that O3-mini, its new model, shows “your thinking process.” Until now the company hid this information to prevent competitors could collect that information to compete with them.
No GPT-5, little of Dall-e 4. In the same session the OpenAi team was asked when the next great model (which is not reasoning), the expected GPT-5 would arrive. Altman said “we don’t have a date yet.” What his colleague said, Kevin Weil, is that they were working on the successor of Dall-e 3, and that “I think the wait will be worth it.” This image generation model was launched about two years ago and since then it has been overcome by many alternatives.
Images | Village Global
In WorldOfSoftware | Openai has taken everything he wanted from the Internet to train his AI. Now accuses Depseek of stealing his data
