Co-Founder & CEO at Thesis
Automating AI research
Thesis is building Darwin, a hill-climbing machine that searches vast combinatorial spaces of experiments and uses the outcome of each one to choose the next. The point is to make machine learning research compound instead of running as a sequence of manual attempts. Sergio worked on AI R&D at Google X, Nvidia and Stanford's AI Lab, published at NeurIPS and ICML, and worked with Chelsea Finn and Andrew Ng. He holds a Stanford BS in math and computer science and an MS in statistics. Thesis (YC F25) is state of the art on OpenAI's MLE-Bench, reached in one month on $10k of compute, ahead of research teams at Microsoft, Google, Meta and Baidu.
Sergio Charles is the co-founder and CEO of Thesis, an AI research lab automating AI R&D. He worked on AI R&D at Google X, Nvidia and Stanford's AI Lab, has published at NeurIPS and ICML, and worked with Chelsea Finn and Andrew Ng.
A Y Combinator F25 company building Darwin, a machine that runs and directs its own machine learning experiments, using each result to pick the next one. Its stated mission is to make state-of-the-art AI discovery 10 times faster and 10 times cheaper.
Thesis is state of the art on OpenAI's MLE-Bench, which tests how well AI systems train machine learning models autonomously. The company reached that in one month with $10k of compute, ahead of research teams at Microsoft, Google, Meta and Baidu.
Sergio wrote in 2021 about a future where AI could create its own algorithms. Foundation models made that plausible, and Thesis is the attempt to build the system rather than wait for one.
Biology and materials science first, with the longer goal of a system that discovers the next Transformer or the next AlphaFold.
His brother Luigi Charles, who previously built Sphere, a cross-border payments company valued at more than $250M in its Series A.
