OpenAI is in trouble? The core team is "falling apart", GPT-5 "will not be seen in October"
Written by Zhang Yaqi, Wall Street News
Recently, OpenAI Seems to be in some trouble.
This high-profile AI The company has been constantly reporting "internal troubles", with core team members leaving one after another and constant senior personnel changes. It has also officially announced that the format of the developer conference in October this year will be changed and GPT-5 will not be announced.
The successive bad news can't help but make people worry:AI Is the Pioneer losing its shine?
"Talent crisis"? Core team members left one after another
Greg Brockman is the president and co-founder of OpenAI.
Today, he announced that he will take a long vacation until the end of the year, saying that "this is the first time I have relaxed since co-founding OpenAI 9 years ago."
He is a key figure in transforming AI research results into actual products in ChatGPT, and is expected to miss events such as Developer Day in October this year.
John Schulman, another co-founder, has joined competitor Anthropic, the Claude team. He led the alignment/post-training team in a series of work from GPT-3.5, GPT-4 to GPT-4o, working to improve model performance.
Today, he also spoke on X to respond to the issue of leaving. He said that OpenAI is the first and only company he has worked for outside of internships. The choice to move to Anthropic stems from his desire to study AI alignment more deeply, that is, the practice of aligning AI with human values.
In his reply, Sam Altman thanked Schumann for his contribution and recalled the first time the two met in 2015.
Chief scientist Ilya Sutskever, one of the inventors of RLHF Jan Leike, and other important members of the company have chosen to leave and establish new startups.
Apple Brother, a well-known OpenAI insider whistleblower account, said: The team is exhausting itself every day. Now the leadership has become a shell, you should just release the product as soon as possible.
But can OpenAI really release its product on schedule?
GPT-5 is unlikely to be seen in October, will OpenAI’s first-mover advantage be lost?
Unlike previous years, this year's OpenAI Developer Conference will be held in a more low-key manner.
The company confirmed that this year's developer conference will focus on developersCommunityDuring this period, it will update its interactions and APIs, and will not release its next major flagship model GPT-5.
“We are not planning to announce our next model at DevDay,” an OpenAI spokesperson said. “We will be more focused on educating developers about what is available and showing developers what is available.”Xiaobai NavigationhairCommunity's story.
OpenAI's DevDay events this year will be held in San Francisco on October 1, London on October 30, and Singapore on November 21.
Tech media techcrunch believes that in recent months, OpenAI has taken more incremental steps, including focusing on optimizing the performance of existing models GPT-4o and GPT-4o mini. As it becomes more difficult to obtain high-quality training data, OpenAI's technological lead in the generative AI competition may weaken.
According to data from Originality.AI, 35% of the world's top 1,000 websites blocked OpenAI's web crawlers, and about 25% of "high-quality" data was restricted. Research organization Epoch AI predicts that if this trend continues, developers will face data shortages between 2026 and 2032.
To address this challenge, OpenAI has signed expensive licensing deals with publishers and data brokers and developed techniques to improve its models’ reasoning abilities, especially on mathematical problems. Chief Technology Officer Mira Murati has promised future models with “PhD-level” intelligence.
OpenAI revealed in a blog post in May that it had begun training the next “frontier” model, but faced enormous pressure to deliver and high R&D costs.
OpenAI also faces multiple controversies, including using copyrighted data for training, restricting employee nondisclosure agreements, and effectively excludingSafetySlower product cycles may have the beneficial side effect of eliminating AI in the pursuit of more powerful generative AI techniques. SafetyPriority of work.
The article comes from the Internet:OpenAI is in trouble? The core team is "falling apart", GPT-5 "will not be seen in October"
OORT's vision is to build an ecosystem with trust at its core, ensuring fairness and inclusiveness of AI technology. OORT's Vision: Providing Trustworthy AI to Humans Since 2023, AI technologies such as ChatGPT, Google Gemini, and OpenAI Sora have had a significant impact on us. However, people are concerned about how they receive...
相关文章
