Aster: Autonomous Scientific Discovery
over 20x Faster Than Existing Methods · 2026
Read Aster: Autonomous Scientific Discovery by over 20x Faster Than Existing Methods (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-07040.txt.
Description
Emmett Bicker offers Aster, an AI agent that does scientific discovery on its own. The user hands it a task, a starting program, and a script that scores the program. Aster then improves the program step by step, and the paper reports that it needs far fewer rounds than rival frameworks to reach new discoveries. Fewer rounds matter most where each scoring run takes hours, as in long machine learning training jobs.
Aster was tried on five fronts: a math overlap problem, a GPU kernel tuning job, a single-cell denoising job, a neural activity prediction job, and a small-model speed contest. The paper reports top results in all but one case. In the remaining case, on ZAPBench, it matched the best human result while using under one one-hundred-ninetieth of the compute. Users can reach Aster through a web page and an API.
Catalog record
| CREATOR | over 20x Faster Than Existing Methods |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 10 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-07040