Scientists have trained machine learning software to classify supernovae without the traditional use of spectra. The project — the first to use real supernovae data to inform its artificial intelligence — is 82% accurate. Currently, scientists take spectra of 10-percent of the ~10,000 supernovae discovered each year. When the Rubin Observatory goes online, only 0.1-percent of the expected supernovae discoveries will be further studied without the new software.