Anthropic's Claude Science Builds First Full UV Sky Map

Anthropic's Claude Science has successfully built the first complete ultraviolet map of the entire sky. Developed in collaboration with Johns Hopkins astrophysicist Brice Ménard, the system processed massive astronomical datasets to construct a unified view of the universe in the ultraviolet spectrum. Ultraviolet light provides vital astronomical insights, illuminating interstellar dust energized by radiation, revealing rings produced by ancient stellar explosions, and exposing dense gas clouds surrounding newly forming stars.

Capturing a total view of the ultraviolet sky from the ground is impossible because Earth's protective ozone layer blocks UV radiation, forcing astronomers to rely entirely on space-based telescopes. While NASA's GALEX mission previously recorded vast stretches of space, it only covered roughly two-thirds of the sky. Crucially, GALEX omitted bright, active star-forming zones, leaving significant gaps in humanity's observational records of the cosmos.

To resolve these missing segments, Claude Science orchestrated autonomous AI agents that downloaded raw data across multiple historical space missions, calibrated the varying inputs, and stitched them together into a singular dataset. The platform then applied an image inpainting algorithm, which evaluates existing observations to accurately infer and reconstruct missing regions. During evaluation tests, predictions generated by the inpainting model achieved high accuracy, exhibiting an average deviation of just ten percent relative to real physical measurements.

The completed ultraviolet map is intended to serve as valuable instructional material for researchers and students. Beyond its immediate astronomical utility, the project demonstrates how modern AI agents can execute tedious data integration tasks that scientists often put off due to time constraints. For domain practitioners, this accomplishment showcases how machine learning can automate complex multi-source data processing, unlocking previously inaccessible scientific insights from legacy space data.