Anthropic Researcher Uses Claude Science to Create First Complete UV Map of the Sky

Brice Ménard, an astrophysicist at Johns Hopkins University and researcher at Anthropic, used Claude Science to produce the first complete map of the sky in ultraviolet (UV) light. This resource addresses a long-standing gap in astronomical data, where previous missions like NASA’s GALEX (2003–2013) skipped bright regions, including the Milky Way plane, to protect detectors.

Methodology and Inpainting

The process involved orchestrating a team of AI agents to gather, calibrate, and merge disparate UV datasets from missions including GALEX, Swift, and FIMS/SPEAR. To fill the roughly one-third of the sky never observed in UV, Claude applied inpainting, a machine learning technique that predicts missing image data based on surrounding context. The model learned the relationship between UV brightness and other wavelengths (visible, infrared, radio) from the two-thirds of the sky that had been mapped.

Accuracy and Validation

The team validated the model by hiding known UV data and asking Claude to predict it. After refinement, the model estimated the hidden data to within 10% of the real measurements, a difference described as almost imperceptible to the human eye. The final map also incorporates UV light estimates from over 100 million individual stars, inferred from ESA Gaia visible light data.

Collaboration and Iteration

The project required more than a dozen successive versions of the map. Ménard noted that Claude initially missed artifacts caused by leftover atmospheric glow in GALEX observations. After identifying the issue, Claude corrected the glow across all 38,000 GALEX observations within a couple of hours. The collaboration allowed Ménard to complete a high-value educational resource without diverting significant time from his primary research.