Last month, Pavel Rabtsevich—a 28-year-old product manager living in Spain—was hunting for planets. He didn’t have any formal training in astronomy or even a telescope. What he did have was an internet connection and subscriptions to multiple AI tools.

He’d been combing through public astronomical data captured by the Transiting Exoplanet Survey Satellite (TESS), a space telescope operated by MIT astronomers and launched by NASA in 2018. As its name suggests, TESS’ main purpose is to help astronomers find exoplanets (planets outside our home solar system) by measuring changes in stellar luminosity; if a star’s brightness dips at regular intervals, it could mean there’s a planet orbiting it and temporarily blocking some of its light from reaching our telescopes.

The data Rabtsevich was analyzing was taken from a narrow sliver of TESS’s full range, captured from November of last year to early January. He handed the dataset—comprising measurements of over 126,000 stars, according to Rabtsevich—to TypeSafe’s Jev, Anthropic’s Claude Code, and OpenAI’s Codex, prompting the AI systems to look for patterns of dimming that could indicate exoplanets, and iteratively narrowing down a list of candidates. “I didn’t take the models’ first answers,” he told Gizmodo. “I pushed them, round after round, to improve their own methods, and checked what they gave me against other models. “I decided which tests to run, judged whether a result was strong enough, and chose what to keep.”

He eventually homed in on TIC 4206066, a star a little smaller than our own Sun and located roughly 116 light-years from Earth.

Source link