news

highlights:

according to people familiar with the matter, the sudden departure of openai chief technology officer mira murati at the end of last month attracted widespread attention in the industry. although she mentioned in her resignation statement that she wanted to "carry out her own exploration," according to people familiar with the matter, the reasons behind it are complicated.

it is reported that there is a dispute within openai about whether to release the o1 model last month. some employees believe the model isn't quite ready for the market yet, but to solidifyOpenAIexistaia leader in the field, a group of people led by ceo sam altman insisted on releasing the o1 model. this reflects the tension between the company's internal research team and the commercial team, with the former more concerned about product safety and risk control, while the latter is eager to bring products to market to achieve profitability.

as the company's cto, mulati is often at the center of this tug-of-war, dealing not only with disagreements between the research and technology teams and the commercial teams, but also with employee criticism of company co-founder and president greg brockman. greg brockman's leadership style and ongoing dissatisfaction over his role in an internal power struggle last november. the pressure eventually took its toll on mulati, leading to her abrupt resignation last wednesday.

on the same day, chief research officer bob mcgrew and vice president of research barret zoph also announced their decision to leave. this chain reaction further triggered external concerns about the stability of openai. sexual questioning. the resignation of three senior managers on the same day, coupled with the departure of many senior researchers and employees in the past six months, has caused openai's stability to face unprecedented challenges.

to improve its appeal to investors, altman has revealed to employees that openai plans to reorganize its current intricate structure. in openai’s current structure, a non-profit foundation serves as the parent company and controls its profit-making arm. after the reorganization is completed, the for-profit division will no longer be under the control of the not-for-profit parent company.

rushed development and security testing

in the fierce competition in the artificial intelligence industry, openai is eager to demonstrate its technical strength and accelerates the testing and release of the latest artificial intelligence model gpt4-o. according to insiders, the launch process of gpt-4o was extremely hasty, and the security team had to work overtime under time constraints and completed the security assessment in only nine days. although the team requested more time, management insisted on releasing as planned. the reason is that google plans to demonstrate an ai model with similar functions at the i/o conference, and openai hopes to seize the opportunity. although security testing has not been fully completed, preliminary assessments believe that gpt-4o can be released safely. however, subsequent research found that the model's capabilities may have exceeded safety thresholds, raising concerns about potential abuse. openai denied this, insisting that the release decision was based on internal security processes and that gpt-4o underwent a thoughtful security assessment.

in addition, another openai model o1 also faces similar problems. o1 performs well on logic and math tasks and is seen as an important step toward artificial general intelligence (agi). however, the release of the model sparked controversy within the company, with some employees viewing its performance as inconsistent and insufficiently safe. co-founder wojciech zaremba also mentioned conflicts during development. but altman promoted the rapid release of the o1 model in order to demonstrate openai's technological leadership to potential investors. although there were objections from the team that o1 was not ready yet, the objections were ignored. currently, o1 is still undergoing fine-tuning to improve the safety and usefulness of its responses.

regarding the safety of o1, members of the safety and security oversight committee of the openai board of directors stated that the release of the model followed strict evaluation and security measures. still, some experts have expressed concerns about o1's persuasiveness and other security risks. finally, despite the pressure and controversy, a spokesperson for openai still stated that the company is confident in the security assessment of the model and emphasized that hundreds of millions of people have used gpt-4o safely since its release. mulati's spokesman denied that pressure related to a rushed model release or management issues played any role in mulati's decision to resign. "she has worked closely with her team and after a huge win, feels now is the right time to leave," the spokesperson said.

loyalty remains an unsolved mystery

within openai, the controversy over the model release appears to be just part of the strained relationship between altman and mulati. people familiar with the matter said mulati's behavior during altman's brief ouster raised questions about her loyalty. when altman was fired in november, employees had expressed doubts about mulati's loyalty. she was appointed interim ceo by the old board of directors and learned about the dismissal 12 hours before it was announced, while other executives only learned about it when it was announced. some ultraman supporters believe that mulati should have informed ultraman in advance. mulati promised to push for altman's reinstatement after being named interim ceo, but days later she was replaced by emmett shear. altman resumed his position as ceo five days later, and murati returned to his position as chief technology officer.

it is reported that mulati expressed concerns about his leadership style to altman and the board of directors, which may have influenced the board's decision to fire altman. although she denied proactively contacting the board, she admitted providing feedback about altman when contacted. mulati's actions raise further questions about altman's loyalty, which some investors may be unhappy with. in addition, mulati also had to deal with friction caused by brockman, who was known as a workaholic who often joined projects at the last minute, putting pressure on employees. mulati had to step in to try to control the situation or calm employees. eventually, brockman was advised to take a leave of absence, which he later announced on social media. a spokesperson for openai said brockman's decision was voluntary and he will return to the company at the end of the year.

changes in corporate culture

openai has undergone a cultural and structural shift over the past 10 months, which began with altman's brief departure. the departures of co-founders ilya sutskever and john schulman, as well as a mass exodus of security researchers, signaled a shift in the company's focus. after voting in favor of firing altman, sutskofer chose to leave and founded a startup called safe super intelligence, focusing on the security control of future artificial intelligence systems. senior researchers jan leike and schulman have also joined competitor anthropic in expressing concerns about openai's productization trend and neglect of artificial intelligence safety.

as the number of employees has doubled, openai's recruitment strategy has also changed. most of the new employees come from large technology companies rather than traditional artificial intelligence research fields. these new hires focus more on product management and sales than pure research. this change resulted in a significant shift in company culture, moving from research-focused discussions to product development and market deployment.

altman's plan to overhaul the company's structure, freeing nonprofit foundations from control of the for-profit arm and lifting limits on investor earnings, has unsettled some former employees. they worry the changes could weaken openai's commitment to security. however, some employees are optimistic about the changes, saying they will reduce internal conflicts and make the company more efficient.

altman now personally oversees the technology team, signaling that he will be more directly involved in the company's technology development. although openai has lost many of its top executives, altman still needs to prove that the company has enough talent to sustain its success and avoid further internal turmoil. it remains to be seen whether these changes can lead openai down a more commercial and product-oriented path while maintaining its commitment to ai safety. (compiled by tencent technology/wuji)