Łukasz Kaiser
Łukasz Kaiser is a Polish computer scientist who helped light the fuse on the modern era of large language models — and then, unlike almost everyone else who lit that fuse, stayed to watch it burn. He was born in Wrocław and studied computer science and mathematics at the University of Wrocław before defending his doctorate at RWTH Aachen in Germany. That early career belonged entirely to pure Logic: his PhD thesis centered on algorithmic model theory, exploring the interplay between logic and computability in automatic structures, work that won him the E.W. Beth award for outstanding dissertations in logic in 2009. He then held a permanent research post at CNRS in Paris, a logician's life that looked nothing like what came next.
Kaiser joined Google Brain in 2013, moving from the French National Center for Scientific Research, and turned toward Neural Machine Translation, Sequence to Sequence Learning, and Attention. He co-built TensorFlow and led Tensor2Tensor, the library that included the reference implementation of the Transformer. That work culminated in Attention Is All You Need (2017), where Kaiser joined Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan Gomez, and Illia Polosukhin in discarding Recurrence and Convolutions in favor of pure Self-attention — birthing the Transformer. The paper's own credits note that "Lukasz and Aidan spent countless long days designing various parts of and implementing tensor2tensor, replacing our earlier codebase, greatly improving results and massively accelerating our research."
Where his seven co-authors went on to found companies, Kaiser made a different choice, leaving Google in 2021 for OpenAI and becoming the only scientist among the "Eight Transformer Pioneers" who has not started a business, choosing instead to remain at the research frontier. There he worked on GPT-4's long-context capabilities and became a research lead on the o1 and o3 reasoning models, extending Chain-of-thought prompting into a new training paradigm.