What is claimed

The Decoder reported on October 9 that Anthropic’s Claude Science mapped the entire sky in ultraviolet light for the first time. The report says the project is meant to show that AI can take on tedious data work researchers would otherwise not get to. Johns Hopkins astrophysicist Brice Ménard describes the process on the Anthropic website. This article did not open that page separately.

Ultraviolet light shows dust lit by starlight, such as clouds around young stars or rings left by stellar explosions. A complete ultraviolet map did not exist before, because the ozone layer blocks ultraviolet light and it can be measured only from space. NASA’s GALEX mission covered about two-thirds of the sky and skipped bright star-forming regions.

[1]
An unfolded paper map with hills, water, and one red ring, and no writing
The gouache shows a paper map with one red ring and no writing. It stands for the assembled map, and it is not an observatory photograph., AI-generated illustration, not a news photograph

The data and the error

Claude Science coordinates agents that downloaded data from several space missions, calibrated it, and merged it. Missing areas were filled by inpainting: a model uses the data it already has to reconstruct the gaps. In tests, the predictions averaged about a ten percent deviation from the actual measurements.

The report says the map is meant to be teaching material. Ménard suspects that many scientists have been putting off similar projects that AI might now make possible. That is his suspicion, not a list of finished projects.

The ten percent figure is an average deviation in the tests. It does not mean every point on the map is off by only that much. The bright regions GALEX skipped are exactly the regions filled in this way.

[1]