Nach Ilya tragen zwei Nach-1990er die Kernforschung von OpenAI.

After Ilya, who is the person supporting OpenAI's research?

On the eve of the release of GPT-5, Altman publicly introduced his two capable lieutenants born in the 1990s.

Mark Chen, the Chief Research Officer, was on Mark Zuckerberg's first list of recruitment targets. "He rejected Mark Zuckerberg's offer of a $1 billion salary."

Jakub Pachocki, the successor to Ilya, is the new Chief Scientist at OpenAI. He was once praised by Altman as "one of the most outstanding minds of our generation."

It is these two people who support the core research at OpenAI. Chen is responsible for building and managing the research team; Pachocki is responsible for formulating the research roadmap and establishing the long-term technological vision.

In addition, they also revealed many details, such as why the Superalignment Team was disbanded and how they usually collaborate. At such a special time, when Mark Zuckerberg is frantically poaching talent and on the eve of the release of GPT-5, it's not ruled out that Altman has some special considerations, perhaps to reassure everyone.

Netizens joked: The partnership is so strong that even ChatGPT calls them mom and dad.

Mark Zuckerberg has just updated his list.

However, how do you know that Mark Zuckerberg didn't act early but ended up with such a result. (Doge)

Two Pillars of OpenAI Born in the 1990s

These two pillars of OpenAI have many things in common. They were both born in the 1990s (Mark Chen in 1990 and Jakub Pachocki in 1991), have experience in Olympiad competitions, are senior employees at OpenAI, and have participated in core projects including the GPT series.

Next, let's get to know these two people better.

Mark Chen

I believe those who have been following Mark Zuckerberg's talent poaching should be very familiar with him. He was probably Mark Zuckerberg's first target on the list and also the inspiration for Mark Zuckerberg's "increased investment in talent."

Mark Chen is the Chief Research Officer at OpenAI. He once led the creation of DALL - E and participated in the development of GPT - 3 and GPT - 4. He built an early version of the model parallel strategy for GPT - 3, added image recognition capabilities to GPT - 4, and also led the development of the coding model Codex.

Last spring, Mark Zuckerberg had a casual chat with Mark Chen. The main content was that Mark Zuckerberg asked Mark Chen for advice on how to improve his company's generative AI department.

Mark Chen didn't think much of it at the time, so he seriously advised Mark Zuckerberg to increase investment in talent. As a result, it was like opening Mark Zuckerberg's Ren and Du meridians, and the craziest talent war in Silicon Valley history was launched.

At that time, Mark Zuckerberg invited Mark Chen, who inspired him, and asked: Would you consider joining Meta, and what would it take to convince you?

Later, it was reported that the offer seemed to reach $1 billion.

But Mark Chen, who had been at OpenAI for 6 years, declined without much consideration and said that he was very happy working at OpenAI.

He graduated from MIT with a degree in computer science. Before joining OpenAI, he worked as a quantitative researcher at Wall Street companies Jane Street Capital and Integral Technology LLC, developing machine learning models for futures trading.

Jakub Pachocki

He joined OpenAI in its third year of establishment, and Ilya was his work mentor. Now he has been at OpenAI for 8 years and is 34 years old this year. He was in charge of the AI application project for Dota and has served as the head of the inference team and the head of the deep - learning science team.

In 2024, after his mentor Ilya left, he was promoted to Chief Scientist, responsible for overseeing projects such as GPT - 4 and o1. He was once praised by Altman as "one of the most outstanding minds of our generation."

His work experience is relatively simple. After completing his post - doctoral research at Harvard University, he came to OpenAI. Before that, he pursued a doctorate at CMU and obtained his doctorate in just three years. He graduated from the University of Warsaw for his undergraduate studies.

In addition, he also has the identity of a competitive programmer.

During high school, he was shortlisted for the IOI competition finals six times and won a silver medal in 2009. In 2012, he represented his university in the International Collegiate Programming Contest, and his team won the gold medal and ranked second overall. In the same year, he was also the champion of Google Code Jam.

These Two People Support OpenAI's Core Research

If Altman is the public - relations face of OpenAI, then Chen and Pachocki are the behind - the - scenes pillars of the company. After Ilya left, they jointly took on the heavy responsibility of core research.

Their relationship at OpenAI is closer than expected. Chen is mainly responsible for shaping and managing the research team, while Pachocki is responsible for formulating the research roadmap and establishing the long - term technological vision.

However, their roles are not fixed. Chen described them as researchers. As long as they see technical details that can be corrected, they will immediately take action and cooperate with each other.

Perhaps this tacit understanding comes from their participation in international coding competitions. Chen is also the coach of the US Computer Olympiad team, which also influences OpenAI's strategic deployment of actively participating in international competitions.

Recently, OpenAI won the second place in the global programming competition AtCoder, only second to the human contestant Psyho. This result has broken through the human performance level and is of great pioneering significance. In addition, OpenAI has just won a gold medal in the IMO, achieving progress in complex reasoning problems.

Interestingly, Psyho is both OpenAI's number - one competitor and a former employee of OpenAI. Pachocki is also friends with him. They used to participate in competitions together. Now OpenAI's LLM is trying to surpass this mountain in the programming world, just like AlphaGo and Lee Sedol back then.

This is also the common view of Chen and Pachocki. Only by actively challenging the top humans in mathematics and coding can the model make rapid progress, because mathematics and coding are the keys to a more powerful general model.

For example, Psyho also gave a very novel solution in this competition, which is a solution the model has never seen before. And this just gives the model an opportunity to improve its knowledge.

Pachocki said:

Programming and mathematics are actually about creativity, coming up with novel ideas, and connecting ideas from different places.

Indeed, OpenAI has been rapidly releasing new products, from GPT - 4 to the o - series inference models, and then to the upcoming GPT - 5. OpenAI has always maintained this momentum of innovation.

As for how OpenAI can balance open - ended research and product development at the same time, Pachocki said that they are always trying to build general artificial intelligence. But in the process of moving towards AGI, the new technologies that emerge may become a side - line product.

OpenAI is always enthusiastic about building AGI. What they are doing now is to understand and break through its technological bottlenecks, such as what it can do and what it can't do.

The current state - of - the - art inference models can decompose problems and know a lot of knowledge, but they can't really connect knowledge like humans and are still in the early stage of the reasoning paradigm. OpenAI is trying to answer this question.

They believe that the scaling law has not completely collapsed in the construction of this reasoning paradigm. As long as they find the right way to build the model or select available data for training, they can approach the bottleneck of AGI.

Of course, the key lies in the model's autonomous time, that is, how the model can complete autonomous research on complex problems within an effective time without stagnation.

Chen and Pachocki attach more importance to the practicality of AGI as a tool and believe that development should be gradual. However, Ilya's view is completely different from theirs. Ilya once said in an interview that AGI will be a milestone that changes the world.

Therefore, Ilya hoped to shift from designing better models to controlling models. So he once established the Superalignment Team within OpenAI, aiming to invest one - fifth of OpenAI's resources to control super - intelligence.

But after Ilya left the company, the Superalignment Team was quickly disbanded.

When talking about the reason for the disbandment, Chen and Pachocki believe that the so - called alignment research has been integrated into OpenAI's core business. The model must work as expected