(Kaileh57)
I lead Sculptor AI, my independent research organization. My work focuses on machine learning and uncertainty, including simulation-based inference for brown-dwarf atmospheres.
I’m an incoming computer science undergraduate at the University of Washington, class of 2030. My research so far has been through Sculptor AI.
I train neural posterior estimators to account for uncertain molecular opacities in brown-dwarf spectra. On perturbed simulated spectra, opacity marginalization improves nominal 90% C/O interval coverage from 61.4% to 89.4% in matched three-seed ensembles.
I write about model behavior, AI safety, and evaluation on my blog. I also gave the TEDx talk “The Cracks in AI Safety.”
My hardware projects include an RP2040 development board (datasheet) and an NFC PCB business card.
I also make games and small interactive experiments: Graftmancer, Arcade Multiplayer Racing, and falling squares.