Source: The Wall Street Journal

With most of the founding team leaving, is OpenAI on the verge of falling apart?

On Wednesday night, September 25th, OpenAI's Chief Technology Officer Mira Murati resigned, followed by Chief Research Officer Bob McGrew and Vice President Barret Zoph's subsequent departures.

On Friday, September 27th, The Wall Street Journal reported that the event of "OpenAI planning to transition to a for-profit entity" is causing the company to split apart - OpenAI's original mission was to develop AI for the public good, a mission that contradicts the introduction of for-profit products. This undoubtedly marks a significant shift for OpenAI. At its founding in 2015, OpenAI clearly outlined the company's vision: to develop AI technology to "benefit all of humanity, without being constrained by financial return requirements".

This year, the departure of key scientists at OpenAI, Ilya Sutskever, John Schulman, and Jan Leike, is also related to the company's change in operational philosophy.

Simultaneously, internal power struggles among executives, work fatigue, and demands for higher salaries from key employees have further impacted OpenAI.

OpenAI is about to become a commercial company, causing intense internal ideological conflicts.

Since Altman reassumed the position of CEO of OpenAI in November of last year, OpenAI has gradually evolved into a more typical commercial company - the number of employees increased from 770 in November of last year to the current 1700; the company appointed a CFO and CPO this year; individuals with corporate and military backgrounds have joined the board of directors of the company... OpenAI is also increasingly focused on expanding its product line, with some long-term employees stating that this has affected its research focus.

Some employees believe that given the cost of tens of billions of dollars to develop and operate AI models, such a development trend is financially feasible. AI needs to move out of the laboratory and into the real world to change people's lives.

However, some AI scientists who have worked at the company for many years believe that the influx of funds and the prospect of huge profits have eroded the culture of OpenAI.

Almost all employees agree that research oriented towards benefiting the public and rapidly developing commercial business, concentrated within the same organization, has made OpenAI's development very painful. Early OpenAI employee and current CTO of AI startup Cresta, Tim Shi, said:

"It is difficult to achieve both at the same time. A product-first culture is fundamentally different from a scientific research culture, you have to attract different types of talents, perhaps you are building a different type of company."

Founding team departing

In May of this year, OpenAI's co-founder and highly respected scientist Ilya Sutskever resigned, followed shortly by Jan Leike, who co-led the safety team with him, joining the competitor Anthropic. Senior management at OpenAI are concerned that their departures may trigger a larger exodus, and are working to persuade Sutskever to return.

Sutskever told former colleagues at OpenAI that he was seriously considering returning, but shortly after, Brockman called to inform Sutskever that OpenAI had withdrawn the invitation for him to return because senior executives were having trouble determining Sutskever's new role and how he would collaborate with other researchers.

Shortly after, Sutskever founded a new company, focusing on developing the most advanced ai without getting distracted by product releases. The company, named Secure Super Intelligence, has raised 1 billion dollars.

On May 17, Leike posted on X platform:

"I have long disagreed with the OpenAI leadership on core priorities in the company, and we have finally reached a turning point... In the past few years, a shining product has overshadowed the culture of safety and processes."

This spring, OpenAI developed a new model GPT-4o, and researchers were asked to conduct more comprehensive security tests than originally planned, but only given nine days – executives wanted to launch 4o before the Google annual developer conference to attract competitors' attention. Security department staff worked 20 hours a day, with no time to check their own work.

Preliminary results based on incomplete data show that GPT-4o is safe enough for deployment. But after the model was released, individuals familiar with the project said that subsequent analysis revealed that the model exceeded OpenAI's internal standards in terms of persuasiveness – persuasiveness is defined as the ability to create content that persuades people to change beliefs and engage in potentially risky or illegal behavior.

The security team raised this issue with OpenAI executives and worked on resolving it. However, some employees were frustrated with this process, believing that if the company spent more time on security testing, the problem could have been resolved before reaching the users.

Co-founder and top scientist John Schulman told colleagues that he was frustrated with the internal conflicts at OpenAI, disappointed in the failure to successfully bring Sutskever back, and increasingly concerned about the diminishing importance of the company's initial mission. In August, Schulman resigned and joined Anthropic.

On Wednesday, Chief Technology Officer Murati announced her resignation on X platform, stating in the letter that she was leaving OpenAI to make time and space for her own research. Subsequently, Chief Research Officer Bob McGrew and Vice President Barret Zoph also resigned. Shortly after Murati's resignation announcement, OpenAI revealed the news of transitioning to a for-profit company.

At this point, only two of the 11 senior members of the OpenAI founding team remain at the company.

Executives are fighting for territory, experiencing work fatigue, and core employees are demanding higher salaries.

Last autumn, members of the OpenAI's board of directors had enough of CEO Altman's management style, such as instigating other leaders to compete against each other. These board members are concerned that if not controlled, it may harm OpenAI's ability to retain key researchers and executives.

On Wednesday, September 25, Altman stated in a memo to employees that he will be more involved in the company's 'technical and product aspects,' rather than the 'non-technical' aspects he previously focused on, such as fundraising, government relations, and relationships with other companies.$Microsoft (MSFT.US)$Altman mentioned that the company's CTO previously reported to Murati and McGrew and will now report to Altman.

"I certainly don't pretend that this sudden change in leadership is taken for granted, but we are not an ordinary company.

Meanwhile, many OpenAI employees complain of high work pressure and feel they should be compensated. They complain that the company prioritizes product delivery over its original mission - building safe artificial intelligence systems.

Several employees have expressed that Altman is urging the company team to quickly transform breakthrough results into public products, which has brought pressure to the company. Employees have to work overtime in the evenings and weekends in order to launch products in a short period of time. Employees working with leaders such as Murati and McGrew also mentioned that these pressures have become unbearable.

Editor/Rocky