A faster algorithm for 4×4 complex matrix multiplication
AlphaEvolve found an algorithm that multiplies 4×4 complex-valued matrices using 48 scalar multiplications, improving on Strassen’s 1969 algorithm, which uses 49 in that setting.
A field journal of machine-made science
Not research about AI
Buffio is a field journal for scientific findings in which an AI system was the researcher: the thing that proposed the molecule, the structure, the theorem, the material, or the planet. It is not a site about research into artificial intelligence.
AlphaEvolve found an algorithm that multiplies 4×4 complex-valued matrices using 48 scalar multiplications, improving on Strassen’s 1969 algorithm, which uses 49 in that setting.
A multi-agent Gemini system proposed original scientific hypotheses. In one biomedical test, drug-repurposing candidates it suggested for acute myeloid leukemia were later shown, in experiments, to inhibit tumor-cell viability at clinically relevant concentrations.
NeuralGCM combined a differentiable dynamical core with learned physics and produced weather forecasts competitive with leading models, as well as climate simulations that tracked observational metrics over multiple decades.
AlphaFold 3 predicted the joint structure of complexes that include proteins, nucleic acids, small molecules, ions and modified residues, outperforming specialized tools on several of those interaction types.
Coscientist, a GPT-4-based system, autonomously designed, planned and executed chemistry experiments in the laboratory, including palladium-catalyzed cross-coupling reactions.
FunSearch discovered new constructions of large cap sets, including a cap set of size 512 in eight dimensions and an improved asymptotic lower bound — reported as the first discoveries on established open problems in mathematics made with large language models.
GNoME discovered 2.2 million crystal structures below the previous convex hull, of which 381,000 are new stable materials. Of the stable structures, 736 had already been independently realized in experiment.
GraphCast produced 10-day global forecasts at 0.25° resolution in under a minute and outperformed the leading operational deterministic system on 90% of 1,380 verification targets, with better severe-event prediction.
RFdiffusion generated new protein backbones and functional proteins — binders, symmetric assemblies and metal-binding sites — that were experimentally characterized, including a designed influenza binder whose structure nearly matched the design model.
A neural network trained on a growth-inhibition screen discovered abaucin, a compound with narrow-spectrum activity against Acinetobacter baumannii. It acts through lipoprotein trafficking and controlled an infection in a mouse wound model.
AlphaTensor discovered matrix-multiplication algorithms that beat known records for several sizes, including a method for 4×4 matrices over a finite field that uses 47 scalar multiplications, improving on Strassen’s two-level algorithm of 49.
ExoMiner, a deep-learning classifier developed at NASA Ames, validated 301 new exoplanets from Kepler data, adding them to the catalog of confirmed planets.
A deep reinforcement-learning controller, trained in simulation and deployed on the TCV tokamak, magnetically controlled a range of plasma configurations, including a two-plasma “droplet,” without further tuning on the machine.
AlphaFold predicted protein structures with atomic accuracy, even when no similar structure was known, and reached experimental quality in a majority of cases in the CASP14 assessment.
A deep neural network identified halicin, a molecule structurally unlike conventional antibiotics, with bactericidal activity against a wide range of pathogens, including in mouse models of C. difficile and pan-resistant A. baumannii infection.
No entries in this field yet.