news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

what happened to openai?

openai, which has been embroiled in a storm of executive resignations, has recently ignited the enthusiasm of the masses.

on the one hand, it was revealed that the post-investment valuation was as high as 150 billion u.s. dollars, making it the most beautiful cub among the global ai unicorns. on the other hand, it lost three technology giants in a row. the only remaining entrepreneurial veteran in office is altman. this is a bachelor commander. the two days of openai were like a world of ice and fire.

‍ image via @blizaine

‍executive resignations are common in any ai company, but what is different is that the "court battle" that occurred at openai last year undoubtedly gave this storm a layer of conspiracy theory.

soon, the foreign media the information exposed the inside story of openai’s employee turnover.

as usual, let’s talk about the conclusion first. openai’s wave of resignations can be roughly divided into two reasons.first, the executives felt wronged, and second, they were not given enough money.

i feel wronged

it is hard to imagine that public opinion could change so quickly on the same matter.

last year, openai ceo sam altman still played the role of "wei guangzheng" in the "palace fight". at that time, most of the media rated the helmsman as the steve jobs of ai, and was even placed in the 2.0 position after jobs was driven away. script.

this year, as a large number of openai executives left one after another, public opinion suddenly turned on altman.

friends who don’t know about this matter can click back to read our report yesterday - is openai not open anymore? company restructuring, high-rise earthquake, but it’s not that bad

to put it simply, in the early hours of yesterday morning, mira murati, the chief technology officer of openai, suddenly announced her resignation.

a few hours after murati announced his resignation, openai chief research officer bob mcgrew and vice president of research barret zoph also announced their resignation plans.

at the time, altman also hinted at the x platform that mira, bob and barret made these decisions independently and amicably.note that the emphasis here is on "independent and friendly".

in fact, the wave of resignations at openai may also have something to do with altman, the person at the helm.

foreign media the information reported that some people who work directly with altman or his deputies believe thataltman tends to exacerbate power struggles within the company.

when company leadership asked for additional hiring or resources, altman would shy away from making these difficult decisions, so much so that other executives, such as current presidents brockman and murati, sometimes had to assume that role.

in may, jan leike, one of the former leaders of the super alignment team at openai, officially announced his resignation.

on the day of his resignation, jan leike posted more than ten "little essays" angrily.the mental journey targets openai and altman, the words are completely filled with disappointment.

yesterday was my last day as head of hyper-alignment, senior leadership and member of the executive suite at openai.

my team has been on a very exciting journey over the past 3 years or so. we introduce instructgpt, the first large language model (llm) trained with reinforcement learning with human feedback (rlhf), publish the first scalable supervised study on large language models, and work on automated interpretability and weak-to-strong pioneering results have been achieved in generalization. more exciting research results will be released soon.

i cherish my team very much. i am incredibly grateful to work with so many great people, both within the super alignment team and with partners outside the team. openai is home to incredibly smart, friendly, and efficient people.

giving up this job was one of the hardest decisions of my life because we must quickly find ways to guide and control ai systems that are far smarter than us.

i chose to join openai because i think it is the best place in the world to conduct this research. however, i had long-standing disagreements with openai's leadership over the company's core development priorities, and our differences eventually grew to the point where they were irreconcilable.

i firmly believe that we should invest more resources and energy into preparing for the new generation of ai models, including security, monitoring, emergency preparedness, security research, robustness to adversarial attacks, ai alignment issues, research on confidentiality, social impact, etc.

these problems are very complex and challenging to solve, and i worry that our current development path will not successfully achieve our goals.

my team has been facing huge challenges over the past few months.at times we have had difficulty obtaining adequate computing resources, making it increasingly difficult to complete this critical research.

creating machines that surpass human intelligence is an inherently risky undertaking. openai is taking on this huge responsibility on behalf of all of humanity. however, over the past few years, safety culture and processes have been marginalized in the pursuit of compelling product development.

it’s long past time for us to take the far-reaching impact of artificial general intelligence (agi) extremely seriously. we must do our best to prioritize preparing for agi. only in this way can we ensure that general artificial intelligence can benefit all mankind.

openai must transform into a general artificial intelligence company that puts safety first.

i want to say to all openai employees: learn to feel about artificial general intelligence (agi). act with a seriousness worthy of what you are building. i believe you can "roll out" the cultural change needed. i'm counting on you. the whole world is counting on you. :openai-heart:

helen toner, one of the driving forces behind the subsequent "firing" of altman, also revealed more inside information to the outside world.

she said that altman likes to "cut first and then tell later" when it comes to company security matters:

on numerous occasions, he provided us with inaccurate information about the formal security processes the company had in place.

this means it is essentially impossible for the board to know how well these security processes are working or what changes may be needed.

as for why altman returned to the board amid the support of employees, helen said:

media reports were wrong, and many employees believed that openai would be "destroyed" if altman did not return to the board of directors.

many employees don’t want the company to fall apart, don’t want to lose their jobs, and need to liquidate their equity.

employees fear opposition to altman and fear of retaliation if altman returns to power

altman has a history of being fired and accused of misconduct

the above information may also allow us to see a more "three-dimensional" altman.

in addition, a person familiar with the matter said that brockman and murati often disagreed on the company's ai development plans.

although brockman, as the company's president, is usually responsible for the company's overall strategic planning and decision-making, he is also involved in specific projects, such as converting early ai research results into practical product prototypes.

this overlapping role of both decision-maker and executor can easily lead to disagreements with the views and plans of murati and his technical team.

at openai's most important conference this year, murati supported the release of the new model gpt-4o.

this is also an important role murati has played in the past, which is to resolve differences between the company's security and product teams before and after product launches.

computing resources are critical to ai research, and former executive mcgrew was responsible for approving researchers' requests for additional computing resources. murati and mcgrew jointly participated in decisions regarding prioritization and project selection of openai research areas.

according to reports, at the meeting after announcing his resignation, mcgrew, who had worked at openai for nearly eight years, told colleagues that he felt tired and revealed that although he had a swimming pool at home, he had rarely had the opportunity to use it because of work.

and some employees said that altman, who loves "pressure," also frantically pushes the company's team to transform breakthrough research results into products as soon as possible, and it is common to work day and night.

faced with such intensity, even executives like murati and mcgrew are feeling exhausted.

nowadays, with the departure of technology giants such as mira and mcgrew, this is also expected to have an important impact on openai's position in the large model competition, especially since openai is in the stage of developing the next generation large language model orion.

not enough money

at the end of last year, openai had only about 770 employees, and as of now, this number has reached more than 1,600.

the rapidly expanding openai is becoming a large company increasingly focused on making money.

in the past, many ai researchers in the industry wanted to work at openai, because it allowed small teams of a few people to explore research ideas that had not yet been proven.

but now, altman is all about making money, making money and making money. launching new money-making products or improving the technology of chatgpt, the mainstay of profits, has become his top priority.

research organization futuresearch once pointed out that openai's annual recurring revenue reached us$3.4 billion. 84% of the revenue comes from chatgpt paid users, while api interface revenue only accounts for about 15%.

as openai's fame and valuation soar, key researchers are also unwilling to generate electricity for love, and naturally hope to receive more remuneration.

according to the report, over the past few years, openai employees and former employees have cashed out a total value of more than $1.2 billion by selling their "profit shares" (that is, the profits that the company gets after making money) to other investors.

as openai's technology continues to advance, some talented employees still hope to get more compensation, and even clamor to leave the company because of money and other reasons.

murati and other executives, such as chief financial officer sarah friar, who took over in june, have been dealing with these issues in the past year.

moreover, competitors such as safe super intelligence, an ai startup founded by ilya sutskever, the former chief scientist of openai, are also recruiting, and openai's technical personnel are obviously in demand.

according to the report, in the face of "popping" by old rivals, in order to retain researchers, openai's leadership can only promise heavy payments and benefits.

in may of this year, it was revealed that openai forced departing employees to sign extremely harsh severance agreements.

this agreement prohibits them from criticizing openai for life, and even admitting that this agreement exists is a violation of the agreement. if the departing employee refuses to sign the document, or violates the document, openai will take back the equity worth up to millions of dollars that the departing employee may have received.

although altman quickly clarified,"it is true, but i don't know about it. it has never been implemented in the past and will not be done in the future.", but the foreign media vox revealed that one of the relevant agreement documents clearly bears altman's signature.

in addition to being a real hammer, this document also slapped altman hard in the face.

the slap-in-the-face farce sparked outrage among openai employees. according to reports, many openai employees not only shared their "disappointment" on their own company's slack, but also expressed their attitudes privately with several company executives.

facing the wave of resignations, altman, who is close to becoming the bachelor commander, at least appears to the outside world to be very calm.

at an event in italy, he deliberately downplayed the significance of the incident, saying the move was a great "transition" for everyone involved and that he hoped openai would become stronger as a result.

in a recent memo to employees, altman said that he was previously responsible for fundraising, dealing with governments, and establishing partnerships with companies such as microsoft and apple.but now he has decided to get more involved in the company's technology development and product production.

and yesterday, he also posted on the x platform:

leadership changes are normal for companies, especially ones that are growing so fast and demanding, but we are not an ordinary company.

i think the reasons mira explained to me (there's never a good time, anything that doesn't come out of the blue leaks, and she wanted to do this while openai was on the upswing) makes sense.