It has been a little over a week since Deepseek upset the AI world. The introduction of its open weight model – apparently formed on a fraction of specialized IT fleas that leaders in the electrical industry – establish shock waves inside Openai. Not only did employees claimed to see indices that Deepseek had “openai” models “inappropriate to create his family, but the STARTUP’s success asked Wall Street to wonder if companies like OPENAI suited to calculation.
“Deepseek R1 is the spoutnik moment of Ai,” wrote Marc Andreessen, one of the most influential and provocative inventors of Silicon Valley, on x.
In response, Openai is preparing to launch a new model today, before its scheduled schedule. The model, O3-Mini, will make its beginnings both in the API and the cat. Sources indicate that it has O1 level reasoning with a 4-level speed. In other words, it is fast, inexpensive, intelligent and designed to crush Deepseek.
The moment has galvanized the staff of Openai. Inside the company, there is a feeling that – in particular that Deepseek dominates the conversation – OpenNAI must become more effective or risk becoming behind its new competitor.
Part of the problem stems from the origins of Openai as a non -profit research organization before becoming a power of profits. A power struggle in progress between research and products groups, according to employees, has led to a gap between the teams working on advanced reasoning and those working on the cat. (Openai spokesperson Niko Felix, says it is “incorrect” and notes that the heads of these teams, the director of products Kevin Weil and the research director Mark Chen, “come together every week and work in close collaboration to align with the priorities of products and research ”.)
Some inside Openai want the company to build a unified cat product, a model that can say if a question requires advanced reasoning. So far, this has not happened. Instead, a drop-down menu in Chatgpt invites users to decide if they want to use GPT-4O (“ideal for most questions”) or O1 (“use advanced reasoning”).
Some staff members claim that although the cat brings the share of the lion of OPENAI income, O1 draws more attention – and computer resources – from leadership. “Leadership does not care about the cat,” explains a former employee who worked (you guessed it). “Everyone wants to work on O1 because it’s sexy, but the code base has not been designed for experimentation, so there is no momentum.” The former employee asked to remain anonymous, citing a non-disclosure agreement.
Openai has spent years experimenting with the strengthening of learning to refine the model which has ultimately become the advanced reasoning system called O1. (Reinforcement learning is a process that forms AI models with a system of penalties and awards.) Deepseek built the work of learning the reinforcement that Optaai had been launched in order to create its advanced reasoning system , called R1. “They have benefited from knowing that learning to strengthen, applied to language models, works,” explains a former OpenAi researcher who is not authorized to speak publicly about the company.
“Learning to strengthen [DeepSeek] Did is similar to what we have done in Openai, “said another former Openai researcher,” but they did it with better data and a cleaner battery. »»
Openai employees say that the research in the O1 was carried out in a code base, called the “Berry” battery, built for speed. “There were compromises – experimental rigor for debit,” explains a former employee with direct knowledge of the situation.
These compromises had a meaning for the O1, which was essentially a huge experience, despite the basic code limitations. They did not make as much sense for the cat, a product used by millions of users built on a different and more reliable battery. When O1 launched and became a product, the cracks began to emerge in the internal processes of Openai. “It was like:” Why are we doing this in the experimental code base, should we not do this in the basic research code on the main products? ” “, Explains the employee. “There was a major decline in internal.”
