Did Sam Altman really say OpenAI would not exist without Polish researchers?
Here is the short version, because the quote keeps getting stretched. In September 2025 Sam Altman published a post about two researchers, Jakub Pachocki and Szymon Sidor, and wrote that without them OpenAI would not be OpenAI. Not "Poland is talented." Not "we have a nice office in Warsaw." Two names. One claim.
Pachocki is now OpenAI's chief scientist. He took the seat Ilya Sutskever left. He is from Gdynia, went through the University of Warsaw, then Carnegie Mellon and Harvard, and joined OpenAI in 2017 when the place was still a small research lab. He led the team that trained an agent to beat world champions at Dota 2. Later he ran the pretraining of GPT-4. Time put him on its list of the hundred most influential people in AI.
Sidor sat in the same high school classroom in Gdynia. Cambridge, then MIT, then OpenAI from 2016. The two of them built a lot of the training infrastructure, including Rapid, the distributed system that made large-scale reinforcement learning actually run. Altman said the company has not yet thrown them a problem they could not solve. That is the kind of sentence a CEO usually saves for a eulogy, not a blog post.
It was never just two people
Wojciech Zaremba, from Kluczbork, is one of the eleven co-founders. PhD under Yann LeCun. Early robotics, then Codex, GitHub Copilot, and the GPT line. Lukasz Kaiser, from Wroclaw, is a co-author of Attention Is All You Need, the 2017 paper that every large language model still sits on. He later moved from Google Brain to OpenAI and worked on reasoning models. Aleksander Madry, an MIT professor, ran OpenAI's Preparedness team, the group that tries to measure catastrophic risk before a model ships. At Google, Jack Krawczyk, born in Wroclaw, led product for Bard and then Gemini.
In the early days, when OpenAI had about fifty people, ten of them were Polish. Five of the seven researchers most closely tied to the o1 reasoning models were Polish too. People inside the company even joked about a "Polish mafia." The joke only works if the density is real.
Wait. Is this just a nice story about immigrants who happened to be in the room, or did they actually change the research path?
They changed the path. The Dota 2 bet was not fashionable. A lot of serious people thought scaling trial-and-error learning that far was a dead end. Pachocki and Sidor pushed it anyway, then reused the same muscle on GPT-4 pretraining and later on reasoning. Kaiser had already written the architecture everyone else trained on. Zaremba helped found the lab. That is not "a Polish accent in the cafeteria." That is the spine of the product you type into every day.
Anthropic does not have the same cluster of Polish names at the very top. That part of the lore is thinner, and it is honest to say so. The concentration is at OpenAI, with a second spike at Google around Gemini, and a third in companies Poles founded themselves, like ElevenLabs. The pattern is not "every frontier lab is secretly Polish." The pattern is that when you open the credits on the systems that actually moved the field, Polish names keep showing up in the jobs that matter.