news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

open ai, which is planning its eighth round of financing, describes a bright future to investors:

in documents disclosed by the new york times, open ai’s active users and revenue have maintained rapid growth. as of august this year, its monthly revenue reached us$300 million, an increase of 1,700% compared to the beginning of the year. in june this year, its monthly active users reached 350 million, more than three times that of three months ago.

according to its forecast, open ai is expected to have annual revenue of approximately us$3.7 billion in 2024. revenue is expected to grow to $11.6 billion in 2025 and is expected to reach $100 billion in 2029.

if the financing is successfully completed, the valuation of the world's third-largest unicorn company will reach us$150 billion.

but behind the increasingly high valuation, this leading start-up company that gathers top talents is casting a shadow - chief technology officer mira and chief scientist ilya have resigned successively, president brockman is "forced" to take a leave of absence, open ai in the end, only sam altman was left in the core four-person leadership team.

light cone intelligent mapping, please indicate when reprinting

from non-profit to for-profit, open ai is destined to turn the ship around and complete the transformation amid skepticism.

why are top talents so frequently lost?

"i made a difficult decision to stay at open ai." last wednesday, an open ai employee left a sentiment on the x platform.

open ai, which has caused personnel turmoil one after another, is shaking the confidence of employees.

after two waves of resignations in november last year and may this year, open ai has just launched the gpt-4o model in full. last friday, it welcomed mira murati, chief technology officer, bob mcgrew, chief research officer, and r&d vice president barret zoph announced the resignation of three employees. in the past two years, more than 30 employees of open ai have left one after another, and only three of the 11 founding team members remain.

"the reason why i want to leave is to create time and space for my own exploration." mira explained the reason for her departure.

Open AI前CTO Mira Murati

it looks peaceful on the surface, but people cannot let go of their suspicions about open ai. throw away issues such as 996 and employee benefits. open ai still has two major disputes that remain unresolved, becoming the sword of damocles hovering over the team:

the first is the concern for safe alignment.

“security culture and processes have given way to shinier products.” in may this year, jan leike, a scientist on the open ai security alignment team, announced his resignation and expressed this emotion on x. in a series of more than a dozen tweets, leike elaborated on his dissatisfaction with open al's disregard for security issues.

now, the problem of open ai’s neglect of the security team seems to still exist, and is gradually coming to the forefront from internal disputes within the company:

just last wednesday, open ai once again staged a "sniper attack" on its strong competitor google: on the same day that google released the gemini family, open ai officially launched the voice version of the product "gpt-4o" that had previously been tested on a small scale. on the table, it is allowed to be used by plus and subscription members. unsurprisingly, open ai successfully stole the attention of users.

but behind this precise stuck point and successful marketing, open ai’s security alignment team is suffering.

according to the wall street journal, the release of gpt-4o was a temporary “benchmarking operation.” within 9 days before going online, the open ai security team was forced to complete the test in a hurry, and the security team worked up to 20 hours a day.

even so, after gpt-4o was fully launched, new problems arose in internal evaluation. subsequent analysis found that the model’s persuasive capabilities exceeded openai’s internal standards for creating content that could persuade people to change their beliefs and engage in potentially dangerous or illegal behavior, people familiar with the matter said.

open ai’s internal decision to turn the ship around and embrace profit-making also made employees feel uneasy.

according to the wall street journal, before co-founder john left, he expressed his concerns to his colleagues: ilya, the former technology backbone, had left, and open ai’s decision to turn to profit made him question the purity of the company.

open ai is becoming an early version of deepmind, providing talent to the industry. at present, some of the employees who have resigned from open ai have chosen to start a new business and start their entrepreneurial careers, while others have gone to open ai's strong competitors. according to statistics, many employees have chosen to join anthropic and google.

after core team members jan leike, john schulman and others joined anthropic one after another, anthropic, which has always been ranked second, is gradually catching up or even surpassing open al.

three months ago, anthropic launched the claude 3.5 sonnet model. according to information disclosed on anthropic’s official website, claude’s score has exceeded gpt-4o in terms of performance in programming, mathematics, visual understanding and other indicators.

in terms of multiple capability indicators, claude 3.5 sonnet surpasses gpt-4o. source: anthropic

a necessary transformation

open ai is a hot top technology company and a top money-burning company.

up to now, open ai has completed 7 rounds of financing, and its financing scale has exceeded 11.5 billion us dollars. among them, microsoft has always been the largest funder of open ai. it has participated in three rounds of financing, one of which injected as much as 10 billion u.s. dollars, setting a record for the largest financing in the ai ​​field.

but the rate at which open ai is burning money is also accelerating. even with projected annual revenue of $3.7 billion, open ai is still a gold-swallowing beast.

first of all, open ai’s employee base is expanding rapidly from 770 people last year to 1,700 people this year.

in addition, in terms of model training, from gpt-1 to gpt-4, as the model parameters gradually expand from hundreds of millions to hundreds of billions, the cost of open ai training is snowballing. in public, sam altman has shared that the training cost of gpt-4 exceeds us$100 million.

under the accumulation of high r&d and training costs, open ai’s income is a drop in the bucket.

according to open al’s internal documents disclosed by the new york times recently, people familiar with the matter estimate that the loss will still exceed us$5 billion, excluding equity and compensation.

as early as july this year, some foreign media also made a not optimistic judgment - if this speed is maintained, the cash on open ai's account will be exhausted in the next year.

with funds about to bottom out, open ai chose to launch a new round of financing again.

last week, open ai’s new financing plan was released, with plans to raise us$6.5 billion, exceeding this year’s record of us$6 billion raised by musk’s ai platform xai. according to current information, this round is expected to be participated by microsoft, nvidia, thrive capital, tiger global, khosla ventures and the uae sovereign fund.

if this round of financing is successfully completed, open ai's valuation will exceed us$150 billion, and it will still be the third most valuable unicorn company in the world, preceded by bytedance and space x.

top 5 open ai ranks third in "2024 hurun global unicorn list" source: hurun report

but faced with a technology company that has been burning cash for a long time and has temporarily limited economic returns, investors' attitudes began to waver. apple made its attitude clear early on. last sunday, foreign media reported that apple had chosen to withdraw from this round of financing.

investors' wavering has intensified doubts about open al. after multiple rounds of financing, open ai faces two expectations - it must not only maintain its leading position in the world, but also provide investors with a more certain future.

in the agreement signed earlier between open ai and investors, the urgency of making money has already been laid. according to reuters, if openai still adheres to the non-profit plan, investors in open ai will have the right to ask the company to return their previous investment. currently, open ai is working on a plan to reorganize its core business into a for-profit company that will no longer be controlled by its nonprofit board of directors.

under heavy pressure, sam altman, the only remaining core member of open ai, had to turn the ship around and shift from a pure research company to the pursuit of profit.

from the perspective of commercialization performance, open ai's most profitable business is still the c-side subscription business. with the rapid expansion of user scale, open ai's revenue has also increased. it is expected that open ai revenue will reach us$3.7 billion in 2024.

the new york times disclosed a document sent by open ai to investors, which showed that as of august, open a1’s monthly revenue reached us$300 million, an increase of 1,700% compared to the beginning of the year. as of june, its monthly active users reached 350 million, more than three times that of three months ago.

although the specific source of revenue has not been disclosed, based on the disclosed number of 11 million subscribers, each user contributes a monthly subscription fee of us$20, and chatgpt’s monthly subscription revenue reaches us$200 million. this is also the main source of revenue for open ai.

the remaining revenue mostly comes from cooperation with third-party developers and enterprises. the document disclosed that more than 1 million third-party partners currently use open ai's technology.

open ai is not satisfied with the current rate of making money. in the document, open ai mentioned some specific plans and predictions about the future, such as increasing the membership subscription price and increasing the subscription price of chatgpt from us$20 to us$44 per month. it expects revenue to reach $11.6 billion by 2025.

in the case of increasing subscription prices, users' willingness to pay and the growth in the number of subscribers are still unknown. but no matter what, open ai is destined to go through this long period of pain.

the next deepmind?

ten years ago, two major figures in the technology world, elon mask and sam altman, gathered at a cocktail party and talked about deepmind, an artificial intelligence company that was in full swing at the time.

sam altman (left) and elon mask (right)

this year, deepmind was acquired by the internet giant google, which is still the number one player in the field of artificial intelligence. the two worry that machines that are smarter than humans will also be dangerous if they are built by profit-seeking companies.

with the same insights, the two joined forces to establish a non-profit artificial intelligence company, and open ai was born.

two years after being acquired by google, alphago released by deepmind shone in a world chess game. alphago defeated world go champion lee jae-seok with a score of 4:1, and has attracted much attention since then.

on the road to agi, the acquired deepmind does not need to worry about the source of funds.

the birth of alphago was built on google's strong capital pool. although deepmind did not disclose relevant cost information. however, in the eight years since deepmind joined google, it has suffered losses for six consecutive years from 2014 to 2019, totaling 1.435 billion pounds (approximately rmb 13.4 billion).

at that time, open ai, which adhered to the non-profit route, was looking for money on its own on the road to agi. after elon mask chose to leave open ai and canceled his previous commitment to invest $1 billion. sam altman found open ai’s largest funder, microsoft, which filled the $1 billion gap.

deepmind, which does not have to worry about making money, has also lost the freedom to make independent decisions.

before joining google, the founders of deepmind tried to remain neutral and asked google to sign an "ethics and safety review agreement", which required deepmind to submit core technologies to the ethics committee for review. but after one meeting, the committee seemed to have never exercised its rights again.

according to foreign media reports, after joining google, deepmind sought independence many times internally, but failed. at this stage, deepmind, which had previously been united, began to lose a large number of employees. it was nicknamed the "whampoa military academy" in the ai ​​industry. some of the employees who left the company started a number of start-up companies.

finally, deepmind officially merged with google's internal team google brain last year to form google deepmind. the merged departments worked together to polish the gemini large model and realize google's plan to catch up with open ai.

after the development of chatgpt, open ai's choices began to run counter to the previous ones - closed source technology, product first and pursuit of profit.

open ai’s ambitions are no longer limited to exploring top technologies. insiders' comment on open ai is that sam is focusing on the product.

starting from the year before last, open ai is no longer limited to the development of the gpt1-4 series, and the development of "shining products" has been put on the agenda. sam altman is trying to turn open ai into a "product-based" company. chatgpt was a successful debut. in just 5 days, it became the fastest product in the world to reach one million users.

while it takes time to explore large-scale model technology routes, open ai has been making continuous moves on the product side. it has successively released products such as gpts store, ai search engine, and end-to-end model gpt-4o.

in terms of product promotion strategy, open ai’s way of sniping at google is simple and effective - to release new products before or on the same day as its competitors’ key launches.

in the nearly half-year product release agenda, open ai is full of firepower against google: first, it launched the vincent video model sora on the day gemini 1.5 pro was released in february, and then released the end-to-end model gpt-4o the day before google’s annual developer conference. , twice successfully stole google’s limelight.

but by fighting google on its own, open ai also paid its own price. although it is not sold to microsoft, open ai has become the property of a commercial organization, causing internal conflicts and brain drain.

deepmind, which was incorporated, independently turned to open ai, a for-profit organization, but ultimately did not escape the pressure of funding.

but for now, will open ai, which at least insists on operating independently, have a better fate than deepmind?

this article comes from the wechat public account "light cone intelligence" (id: guangzhui-tech), author: wei linhua‍‍‍‍‍‍