Scientific discovery has long been one of America's greatest strengths. That's why we're deepening our partnership with the U.S. Department of Energy's #GenesisMission# through new investments in AI infrastructure, scientific computing, engineering expertise, and a collaboration hub designed to help researchers across America's national laboratories, universities, and industry work together more effectively. Our goal is to accelerate AI-driven scientific discovery. From energy and medicine to advanced materials, faster scientific breakthroughs can strengthen America's competitiveness and create lasting value for generations to come. @ENERGY @Microsoft

Our co-authored paper with @OpenAI and leading researchers in genomics and scientific computing is now available as a preprint on bioRxiv. This work explores what becomes possible when frontier AI agents and domain experts come together to build and scale complex scientific software, including HelixForge, built by Minos. Read the full preprint:

My most recent claim is that "FP8 is all you need" for all scientific computing. Two new papers as well as my debate with my great friend Jack Dongarra are covered here: (Revised) Part1 paper is on arXiv: Part2 is on hold at arXiv but you can access it here: Comments are welcome!

Excited to share FloatLib, our verified arbitrary-precision floating-point arithmetic library in Lean. We’ve spent several months trying to bring together the best of both worlds: arithmetic we can prove correct and implementations that run efficiently. We built FloatLib to support verified machine learning and scientific computing, where rounding, overflow, and accumulation can change a program’s result. FloatLib supports IEEE binary and decimal, arbitrary-width posits, P3109, and small ML formats. You can also define your own formats and rounding rules. Each certified software backend comes with a Lean proof that it computes the specified result, including signed zeros and exceptional values. A lot of the work went into making those implementations faster, with lookup tables for tiny formats, machine-word kernels, and limb algorithms for wider arithmetic. They share the same specifications, so each optimization must come with a proof that it preserves the result. We also put FloatLib through extensive numerical checks and speed comparisons with established libraries, including MPFR, Flocq, FLoPS, Berkeley SoftFloat/TestFloat, and the posit libraries SoftPosit and Universal, across a range of formats, operations, and precisions. @Robertljg Project & Paper: Code: