Learning to Discover at Test Time

Mert Yuksekgonul, Daniel Koceja, Xinhao Li, Federico Bianchi, Jed McCaleb, Xiaolong Wang, Jan Kautz, Yejin Choi, James Zou, Carlos Guestrin, Yu Sun · 2026

Read Learning to Discover at Test Time by Mert Yuksekgonul, Daniel Koceja, Xinhao Li, Federico Bianchi, Jed McCaleb, Xiaolong Wang, Jan Kautz, Yejin Choi, James Zou, Carlos Guestrin, Yu Sun (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-2601-16175.txt.

Introduction

Mert Yuksekgonul and coauthors let a model keep learning at test time to find new best marks. Their 2026 preprint asks how AI can find a new top result for a science task. Past test time search prompts a fixed model to try new paths. Here the model trains at test time on that one task. Its train gains skill tied to the task at hand.

Live train lets the model learn from its own tries on the task. Each try gives feed back the fixed path can not use. The team tests if live train beats fixed search. Gains would open a new way to push science tasks. The work frames test time as a train phase.

Catalog record

CREATORMert Yuksekgonul, Daniel Koceja, Xinhao Li, Federico Bianchi, Jed McCaleb, Xiaolong Wang, Jan Kautz, Yejin Choi, James Zou, Carlos Guestrin, Yu Sun
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE193 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-16175

Browse the full library · About the Tower of Babel