Field notes on AI for Science
Landscapes, datasets and teardowns from the Spark Science research notes, written for practitioners.
The Landscape of Scientific Simulation
Simulators across 13 disciplines: how big each sim2real gap really is, how to measure it, how simulation feeds AI, and where the money flows.
51 min read
BenchmarksWho Writes the Test, Who Grades It, and Can the Test Be Trusted?
A supply-chain map of 64 AI benchmarks: who writes them, how harness choice skews scores, and why rubric grading falls short.
55 min read
Live dataGPU Prices: China vs US
A weekly panel comparing H100, H200, B200, A100 and RTX rental and purchase prices in China and the US.
8 min read
AI for ScienceAI for Science Atlas: six disciplines
Six disciplines, one question: where AI has a verifier to trust, and where the evidence is still thin.
18 min read
AI for BioAI for Bio: Where the Value Lands
Five judgements from 147 notes: AI cheapened the cheapest step, biology lacks outcome data, and China makes molecules the US prices.
27 min read