The guy who created LightFM has moved from DeepMind to OpenAI. Maciej Kula. Back in 2015, he was a data scientist at a fashion e-commerce company in London. He single-handedly published a workshop paper called LightFM, which tackled the cold start problem in recommendation systems: even when new users have no activity records or new products haven’t been purchased yet, it could still make accurate recommendations. Eleven years later, it’s still the go-to baseline for many companies working on recommendations. In 2019, he joined Google, working on recommendation systems for YouTube, Search, Play, and Ads, and wrote TensorFlow Recommenders. In the past two years, he shifted focus to post-training for language models, tool usage, and synthetic data, serving as the co-lead for Gemini tools research. In other words, before he left, his work was essentially "teaching models how to use tools." OpenAI is now betting on agents, and this is exactly what they’re missing. Oh, and by the way, in his free time, he’s written machine learning libraries and automatic differentiation in Rust by hand. On GitHub, he only follows 2 people.