The simple Thursday ritual that made OpenAI's engineers into AI researchers
Curated by the Inblix editorial team
For years, the unspoken rule at top AI labs was that engineers stayed engineers and researchers stayed researchers. Cross-pollination was rare. OpenAI decided that was a bug, not a feature, and built a structural fix around it: Learning Day. Every Thursday, employees across the company step away from their main projects to do something that makes them sweat in an unfamiliar technical domain. Not leisure. Not catching up on email. Hard, focused self-study.
The idea came from Wojciech Zaremba, OpenAI’s head of robotics, who drafted the initial guidelines for his team before the concept spread. The rules are deliberately strict to prevent the day from being swallowed by urgent-but-unimportant work. Everyone posts what they learned in Slack, creating a public log that functions as both accountability and inspiration. The company also reimburses books and tutors—mostly for math fundamentals—which it considers a trivial cost relative to the payoff.
What people actually do on a typical Thursday is granular and gloriously nerdy. They reimplement papers like MAML from scratch. They train LSTMs and transformers on the Penn Treebank dataset. They read dense material ranging from Judea Pearl’s The Book of Why to histories of Soviet technology transfer. The reading list alone on a single Learning Day spans a dozen machine learning papers, from Population Based Augmentation to Weight Agnostic Neural Networks. The point is breadth, not immediate utility.
Before this experiment, OpenAI rarely saw someone from a pure software background pick up machine learning skills. Now that kind of growth is, in their words, very common. The model has held up against two obvious failure modes: the day being co-opted for regular work, and the habit leaking into the rest of the week. Scheduling it for the same day across every team creates positive peer pressure. And so far, no one’s treating every day like Learning Day—which, OpenAI notes, would probably signal they’ve lost enthusiasm for their actual job.
💡 Key Takeaways
- OpenAI credits Learning Day with making cross-functional skill growth common—something it says was rare before, and remains rare in industry outside academia.
- The program succeeds in part because of rigid rules: same day for all teams, public Slack accountability, and a clear distinction between hard study and leisure.
- The reading and coding logs reveal a strikingly broad curriculum, from implementing JAX experiments to studying the history of cyber warfare and Soviet tech transfer.
- OpenAI considers book and tutor reimbursements for math fundamentals a negligible expense compared to the long-term returns in employee capability.
Keep reading: See related articles below for more coverage on this topic.
Get smarter about AI
The sharpest AI news, curated daily. Delivered free to your inbox.