Deepseek Openai was fired

It’s been a week since DeepSeek Approved the EI world. The application of the open weight model, is probably trained in part of the specialization chips chips Electricity industry leaders – shocked waves in Openai. Only the employees claimed that Deepseek did not see the instructions to create themselves, but the success of the beginning Wall Street was asked, and said they calculated the reporting of the passengers.

“DeepSeek is the Sputnik’s moment of R1,” Silicon Valley wrote Marc Andreessen, one of the most influential and provocative inventors. In x.

In response, Openai is preparing to advance the scheduled schedule of a new model today. The model will debut in the conversation in both the O3-mini, API. Sources say that this is a 4-level rapid substantiation. In other words, it is designed to crush fast, inexpensive, smart and DeepSeek.

At the moment, the Openai delegation was sinned. Inside the company, especially Deepseek’s conversation has a feeling – Openai should risk falling behind a more efficient or most new opponent.

Part of the issue is caused by Openai’s origins as a non-profit research organization before there is a profit seeker. A power struggle between research and crop groups, the workers, developed, developed thinking and conversation, caused a reef. (Openai spokesman Niko Felix, the leaders of these teams, chen, General Research, Mark Chen, “Each week, records meetings and products and research priorities.”)

Some Openai wants to build a model that will say that the company’s unified conversation product requires an advanced thinking of a question. This has not happened so far. Instead, a menu opened in ChatGTT is to use GPT-4O (“Wants” for most questions “or use O1 (” advanced justification).

When chatting with some employees, O1 more attention and computing resources – when the open income of the conversation is brought to the lion’s share, it is due to the leadership. “Leadership does not pay attention to the conversation,” says a past employee. “Everyone wants to work on O1, because it was sexy, but the code database was not built for experience, so there is no speed.” The former worker wanted to remain anonymous, no indefinite agreement asked.

Openai spent years by strengthening the model to make a model of an advanced system called O1. (Strengthening is a process that prepares AI models with the penalty and premium system. “They say that there is an Openai researcher who has no authority to speak to language models, applied to language models, because they are applied to the language models.

“Accelerator Learning (DeepSeek) is similar to what we do in Openai,” he said.

Openai employees say the investigation into O1 is made in a coding base called the “Berry” stack built for speed. “Experimental for the experience,” the situation is a direct knowledge with knowledge directly.

He gave meaning to O1 with a trading-offs, in fact a very large experience, code database restrictions. For conversation, many meanings did not give a product used by millions of users built on a different, safer pile. When O1 put into operation and a product, the cracks began to appear in Openai’s internal processes. “It was like this, ‘Why do we do this in an experimental code, we should not in the main product research code?” “The employee explains. “It was the main push for the interior.”