news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

New Intelligence Report

Editor: Editorial Department

【New Wisdom Introduction】OpenAI has experienced a series of changes in its core senior personnel. In troubled times, heroes emerge, and a group of up-and-coming talents take the lead.

Since Ultraman was dismissed and reinstated in November 2023, there seems to be a constant power struggle within OpenAI.

Recently, many senior executives have left the company. According to the Financial Times, of the original 11 co-founders of OpenAI, only two are still working at OpenAI.

Apart from Altman, only Polish computer scientist Wojciech Zaremba remains at OpenAI as a researcher.

Greg Brockman, president and co-founder of OpenAI, who has been working alongside Altman, will extend his leave until the end of this year, which is widely seen as a precursor to his imminent departure.

Another co-founder, John Schulman, left the company permanently to join competitor Anthropic, and Peter Deng, vice president of consumer products, also left OpenAI.

According to statistics, nearly 75 core employees of OpenAI have left and founded about 30 AI startups.

OpenAI has become the "Whampoa Military Academy" of the AI ​​industry, and its former employees have set up their own businesses based on the experience and connections they accumulated in the company.

Although this talent output is beneficial to the development of the entire industry, it is undoubtedly a huge loss for OpenAI itself.

Rookie takes over

Although there has been a loss of talent, OpenAI continues to spend heavily to recruit and promote talent, and other people have stepped forward to take over the vacancies.

This kind of scene makes people sigh: "Game of Thrones" happens in reality, "chaos is the ladder to rise".

Here’s a look at OpenAI’s new crop of leaders, based on conversations with former and current employees and other insiders.

Jakub Pachocki——The "genius scientist" who succeeded Ilya

Jakub Pachocki is currently the chief scientist of OpenAI and has been working at OpenAI since 2017.

Judging from Pachocki's past experience and his rocket-like promotion speed within OpenAI, he deserves the title of "genius scientist".

Before joining OpenAI, Pachocki obtained his Ph.D. from CMU in just three years and conducted postdoctoral research at Harvard University for seven months.

In the first five and a half years after joining, Pachocki led the Dota, reasoning and deep learning science teams, all of which are part of OpenAI's transformative research programs.

He played a key role in OpenAI’s development of a Dota 2-playing bot that played against itself to a professional level.

Since 2021, Pachocki has been promoted to research head, research director and other positions. He is the leader in the development of GPT-4 and OpenAI Five, and has been praised for his research contributions in large-scale reinforcement learning and deep learning optimization.

Altman was effusive in his praise for Pachocki, describing him as "undoubtedly one of the greatest thinkers of our generation" and "responsible for the stewardship of many of our most important projects."

Regarding Pachocki's contribution to the GPT-4 project, Altman once commented: "The outstanding leadership and technical foresight demonstrated by Jakub Pachocki are remarkable. Without his contribution, we would not have achieved what we have today."

In a previous MIT interview, Pachocki compared developing a language model like GPT-4 to "building a spaceship" - you have to make sure every part is perfect.

He also mentioned that the basic construction method of the GPT model has not changed much since the first version was released in 2018.

Pachocki officially took over as chief scientist after Ilya Sutskever left in May this year.

Pachocki’s growing influence preceded his promotion, according to current and former employees.

Ultraman thinks highly of Pachocki and has previously assigned him tasks similar to those of Ilya, even though Pachocki should report to Ilya, which has led to tension between the two.

During the "palace fighting storm" in November last year, Ilya and board member Helen Toner jointly launched the "expel Ultraman" action, but Pachocki did not follow Ilya and stood firmly in the camp supporting Ultraman.

However, when Ilya officially announced his resignation, Pachocki's tweet still showed his respect and admiration for his predecessor.

Ilya introduced me to the world of deep learning research and has been a mentor and great collaborator for me over the years. His incredible vision for deep learning is fundamental to what OpenAI and the field of AI are today. I am very grateful to him for the countless conversations we had, from high-level discussions about the future of AI advancement, to deep technical whiteboard sessions. Ilya - I will miss working with you.

Employees said that after Ilya left, Pachocki received support from Altman and has become an important decision maker in the company.

Pachocki was born in Poland and graduated from the Department of Computer Science at the University of Warsaw. Not only did he get his bachelor's degree in just three years and had two internships at Facebook, he also participated in programming competitions many times during high school and college.

Jakub Pachocki (third from left) at the ICPC 2012 competition

According to the Programming Hall of Fame, Pachocki represented the University of Warsaw at the ACM-ICPC World Finals twice, in 2011 and 2012, and won the gold medal in the second year.

In the individual competition, Pachocki's performance was equally amazing: he won a silver medal in the IOI in high school, and has won medals in programming competitions hosted by Topcoder, Facebook, and Google many times.

Among them, the best result was winning the gold medal in the 2012 Google Code Jam competition, with a prize of 10,000 US dollars.

He once mentioned in an interview that programming competitions such as Google Code Jam are more like math homework or solving logic problems, and winning requires extremely high cognitive abilities.

Barret Zoph

Before co-founder John Schulman left, he co-led the post-training team with his only direct report, Barret Zoph.

According to an employee, Schulman was previously responsible for setting the high-level agenda, while Zoph was responsible for the team's daily management and research, ensuring that the project was completed on time and that the basic model could be smoothly deployed to products such as ChatGPT and APIs for developers.

Zoph is highly respected within OpenAI, and now that his boss has left, he has naturally become the sole person in charge of the team.

Zoph graduated from the University of Southern California with a bachelor's degree. During his college years, he joined the Natural Language Group of the USC Information Sciences Institute and worked on statistical machine translation with two professors, Kevin Knight and Daniel Marcu.

After graduating from college, Zoph joined Google Brain and worked as a research scientist for 6 years and 8 months, focusing on training large language models and applying them to various applications.

During his time at Google Brain, Zoph has made many academic achievements. According to Google Scholar, his papers have been cited 62,284 times.

In August 2022, Zoph resigned from Google Brain and joined OpenAI to participate in the creation of ChatGPT. His main research direction is training large sparse language models and AuoML, such as neural structure search (NAS).

At OpenAI’s May conference, Zoph demonstrated GPT-4o’s real-time vision capabilities, allowing GPT-4o to solve math problems in real time using a mobile phone camera.

Mark Chen

Mark Chen has been the director of cutting-edge research since joining OpenAI in 2018, leading a working group under Vice President of Research Bob McGrew.

Chen also showed extraordinary talent at an early age, and won full marks in mathematics competitions such as AMC10, AMC12, and AIME, which paved the way for him to enter MIT.

When he graduated from MIT, Chen received a double bachelor's degree in mathematics and computer science. During college, he interned at Microsoft and Trading, and was a visiting scholar at Harvard University.

After graduating from college, Chen entered the financial field to engage in quantitative research, building machine learning algorithms for stock and futures trading.

In his last company, Integral Technology, Chen had reached partner level.

Currently, he also serves as the coach of the US IOI training team.

Sure enough, people who are good at math can do anything with ease.

Rest in peace, Jim Simons. You are living proof that being really good at math has a positive impact on almost everything else.

Chen focuses on multimodal modeling and reasoning research at OpenAI. He led the team that created DALL·E and the team that incorporated visual perception into GPT-4.

Chen also led the development of Codex and contributed to advances in GPT models, including the development of image GPT, and was part of the 17-person gold medal team for GPT-4o.

At the launch conference in May this year, Chen also took the stage to demonstrate the voice function of GPT-4o.

Chen's influence gradually became apparent during the "palace fight" in November last year. According to a former employee, Chen, Barret Zoph, and post-training researcher Liam Fedus were the main liaisons between the leadership and employees at the time.

For example, the three delivered a joint letter from employees supporting Altman, which became a key link in pushing Altman to reinstate. Most employees in the letter said that if Altman was not reinstated, they would join Microsoft.

Lilian Weng

Lilian Weng is currently the head of OpenAI's security system, mainly engaged in research in machine learning, deep learning, etc.

Weng graduated from Peking University with a bachelor's degree in Information Systems and Computer Science. He went to the University of Hong Kong for a short exchange and then received his Ph.D. from Indiana University Bloomington.

During his doctoral studies, Weng's research areas were complex systems and networks, focusing on social media, social games, human-computer interaction, and complex information network modeling.

Open her Google Scholar profile and you can also see papers Weng has published on memes and social networks.

She has interned in user analysis at companies such as eBay and Mozilla, and then successfully "changed careers" to work in software engineering and data science at Facebook and Dropbox.

Since 2018, Weng has joined OpenAI as a research scientist and, as technical lead of the robotics team, focuses on algorithms for training robotics tasks.

Weng later led a research team for applied artificial intelligence and is currently the leader of the security team.

In July this year, OpenAI transferred Aleksander Madry, the former head of the security team, to a team focused on the basic work of reasoning. According to people familiar with the matter, the security team originally led by Madry was transferred to Lilian Weng.

Weng will now manage teams focused on both long- and short-term AI safety, an organizational decision that has worried some researchers because the incentives for long- and short-term safety could conflict with each other.

It is worth mentioning that the blog articles written by Weng on his personal website are very popular. They are basically long articles of tens of thousands of words, containing both technical content and opinion output, and are a reference for many people in the industry.

References:

https://www.theinformation.com/articles/the-rising-guard-at-openai?rc=epv9gi

https://www.sahmcapital.com/news/content/meet-jakub-pachocki-openais-new-chief-scientist-after-ilya-sutskevers-exit-sam-altman-hails-him-as-one-of-the-greatest-minds-of-our-generation-2024-05-15