Why doesn’t Ilya Sutskever go to work?

Ilya Sutskever was not fired or officially suspended, but he was not seen at his workplace all last week. In other cases, this would be a typical truant, if we were not talking about the chief scientist of OpenAI.

Why doesn’t Sutskever go to the office? It’s hard to say yet. But now it’s easy to guess why Sam Altman was fired. To do this, you need to unwind the OpenAI tangle to the very foundation of the company.


The name OpenAI has firmly entered the media agenda: the company has some of the best generative neural network models on the market, GPT-3.5, GPT-4 and DALL·E-2. In fact, there is no single OpenAI. Various organizations sell subscriptions to ChatGPT Plus and determine their development strategy.

In December 2015, Elon Musk and Sam Altman founded OpenAI is a non-profit organization (Nonprofit). In the first months, only a press release was posted on openai.com. On simple web page with a minimalist design, the goal was stated: to create universal artificial intelligence that is safe and useful for humanity.

Among the co-founders were many eminent scientists in the field of AI. For example, one of the world’s leading experts in machine learning, Ilya Sutskever, an Israeli-Canadian specialist born in the USSR, was appointed chief researcher. Greg Brockman, the former technical director of the Stripe payment system, was hired as technical director.

The projects that OpenAI produced in these early years were non-application-oriented and intended primarily for other researchers. This is a toolkit for developing and comparing reinforcement learning algorithms Gym (April 2016) and platform for training strong AI Universe (December 2016).

OpenAI was tiny at first. Even a year later, in January 2017, there it worked only 45 people. (Today on OpenAI works about 770 people.) Google poured in without giving hundreds of millions of dollars a year to the DeepMind division. OpenAI’s spending was an order of magnitude smaller.

Already in 2016 understoodthat the price of a top developer in the AI ​​industry is comparable to the cost of a football player. However, it was believed that generous donations would be enough to keep OpenAI afloat.

The head of Nvidia personally hands over the DGX-1 Mask for 170 TFLOPS, 2016. Nvidia

As with the idea of ​​OpenAI, there was little that was unique in the name. Almost two years earlier, in January 2014, Slovak entrepreneur Marek Rosa opened an organization with a similar focus and a suspiciously similar name, GoodAI. But OpenAI was distinguished by its scope: it was planned to raise a billion dollars.

They took it in kind from someone: Nvidia gave the DGX-1 supercomputer, Amazon and Microsoft provided free computing power in their clouds. For several tens of millions of dollars, the sponsor could even get a seat on the company’s board of directors. For example, the Open Philanthropy research foundation bought a seat on the board for 3 years for some $30 million. Musk, Altman, Brockman, Sutskever and several other key specialists had permanent seats on the board of directors.

Over three years, we managed to scrape together only $130.5 million in total. And in 2018, from the board of directors came out Elon Musk. Tesla at that time began to work more closely on autopilot, so the entrepreneur had to leave OpenAI to avoid a conflict of interest. At least that’s the story that went public. Where later Sam will tellthat Elon requested full control of the company, was refused and left because of this.

Musk is one of OpenAI’s key backers. TechCrunch analysis showsthat Musk brought in at least $15 million, although he himself names the numbers in $50 million or $100 million. Publicly, Elon said that he would continue to give OpenAI money, but he did not keep his promise.

The need to attract expensive personnel and pay for computing power has not been canceled. Therefore, in March 2019, OpenAI reported on the formation of the commercial division of OpenAI LP. Sam Altman was appointed to the post of leader – at that time he had already resigned as president of Y Combinator.

OpenAI

This is how a conglomeration of various regulatory bodies was born. The design is deliberately built so that all leverage remains with the non-profit branch of Nonprofit and the members of its board of directors. The commercial subsidiary is legally obligated to follow the Nonprofit mission. Additionally, there is a profit limit for investors and employees. All money in excess of the limit (negotiated individually for each partner) is returned to the Nonprofit to be used for the benefit of all humanity.

Even large investors have little control over what they invest in. In July 2019, OpenAI announced about entering into a partnership with Microsoft. The scope of this cooperation increased to multi-billion dollar in early 2023. At the same time, Microsoft remains only an observer without voting rights on the board of directors.

And the board sometimes makes important decisions. When on November 17 the board of directors shifted Altman, Microsoft had to be content public statementsthat the partnership with OpenAI continues, no matter what. At the same time, behind the scenes, Microsoft and other investors asked put Sam back in his place. In a few days threats walk away from OpenAI employees and negotiations involving Satya Nadella, Microsoft CEO, Altman returned to his post.

Altman’s return did not come without changes on the board of directors. Sam introduced the interim board, which did not include Sutskever’s name. Ilya is one of the board members who supported Altman’s dismissal. We were talking even that he played a central role in this decision.

Why was Altman fired? Details of the conflict have not been disclosed, but information is leaking:

1. In July of this year, Sutskever formed the Superalignment team at OpenAI, which focused on managing future super-powered AI systems. The chief researcher spoke about this, for example, in an interview with MIT Technology Review. And this is a problem: Ilya’s media comments only added fuel to the fire of the narrative about the dangers of artificial intelligence, which created risks for OpenAI’s commercial interests.

   In October, Ilya’s job responsibilities at the company were suddenly reduced due to… indicates insider comments to Bloomberg, friction with Altman and Brockman.

   A few days before the removal of the head of OpenAI, a team of researchers presented report on Q* – about a potential candidate for the status of universal artificial intelligence.

   AGI (artificial general intelligence) is the main interest of OpenAI. By this term, the company understands autonomous systems that are superior to humans in basic economically valuable tasks. OpenAI was literally founded to create AGI.

   It is possible that with this report, Sutzkever and/or his minions wanted to shift OpenAI’s focus from extracting revenue from ChatGPT back to their research.

2. Other source calls the second reason for what happened: Sutskever and the board members who joined him feared that Altman wanted to raise investments at the expense of the OpenAI name and change the structure of the organization.

   Indeed, shortly before this Sam had tried find tens of billions for a company producing accelerator chips for AI. The plan is obvious: take a piece of the pie on which Nvidia raised its capitalization is over a trillion dollars. In search of money for this Tigris project, Altman flew a lot to the Middle East.

3. Finally, it’s not just Sutskever. Altman clashed with others.

   Sam is far from sugar: he characterize as a manipulative person. It is said that Sam has a habit of creating unhealthy conflicts between employees. Altman may react sharply to negative feedback.

   For example, Helen Toner, another board member, recently co-authored an academic paper on AI safety that was mildly critical of OpenAI. Altman accused her of harming the organization and initiated her dismissal from the board of directors by inciting people behind her back to lie and manipulate. To make matters worse, this behavior is not the exception, but the norm. the best Chief Executive Officer of the Year for 2023 according to Time magazine.

   After Sam returned, Helen was also removed from the board. She herself refuses to explain why she supported Altman’s dismissal.

In internal correspondence, the dismissal of OpenAI head Sutskever

described

as correct actions of the board of directors, for which the non-profit organization was created. He renounced these words just a few days later: after Sam’s return, Ilya repented and

apologized

.

The OpenAI office is reacting to the news that the old head has been reinstated to his position. @gdb

In an open letter, Sam explains that he holds no grudge, but Sutskever will no longer take part in the board of directors. As the head of OpenAI wrote on November 29, the question of in what form the chief scientist will continue to work in the company has been discussed.

On the occasion of Altman’s return to OpenAI rolled up the party was so intense that the fog generator set off the fire alarm and fire trucks appeared outside the office windows. That day Ilya also experienced how he wrote in a tweet, there was inexpressible joy at Altman’s return, but for some reason at the event was absent.

Moreover, Ilya Sutskever suddenly acquired a lawyer. In response to Business Insider’s request why OpenAI’s chief scientist did not come to celebrate the return of the head of the company, lawyer Ilya Sukho advised don’t attach any importance to it.

Last work week (since December 4), Sutskever did not show up at the OpenAI office in San Francisco. told Two Business Insider sources at the company. No one cut off Sutskever’s access to corporate systems: he is still available on Slack. There was no attempt to erase his presence: the office walls are decorated Ilya’s art. However, without any official decision, Sutskever for some reason does not go to work.

As a source tells Business Insider, Sutskever’s role has always been huge, but over time his responsibilities are being taken over by other employees. The second source claims that a new position is being prepared for Ilya, as they want to find him another role in the company.

May hint at positive intentions photo [размещена до ката] with the caption “❤️” posted by OpenAI co-founder Greg Brockman. In the photo, Brockman and Sutzkever are smiling and hugging as if Sutskever had not been publicly removed from the OpenAI board of directors two days earlier.

However, not everything is so smooth. Last Wednesday, December 6, Sutzkever posted a tweet with the text: “I have learned a lot of lessons over the past month. One of them: the phrase “they will beat until morale improves” is used much more often than it should be.” The tweet stayed up for about a day, but was then deleted.

Ilya did not erase another hint: it is black and white sketch from the same day, which depicts a face with some probably negative emotion.

Citing a source at Microsoft, Business Insider claims that Altman, Brockman and Sutzkever are unlikely to work together because in Silicon Valley it is considered morally unacceptable for co-founders to turn against each other. Concerns are confirmed by one of OpenAI’s former employees, who believes that employees loyal to Altman will have difficulty working with Sutskever.

Insiders describe Sutskever as an intellectually deep person, immersed in his thoughts. Ilya’s style of work is called visionary, and supposedly this “academic” style did not help win the favor of Altman, Brockman and the team of engineers. On the other hand, Sutskever’s contribution to the company is still respected.

The coming weeks will show how the problems in the relationships of these people will be resolved. In other cases, this would be a typical work conflict if a unique AI company was not at stake c

potential

estimated at $86 billion.