TextBFGS: Quasi-Newton Optimization for Discrete Executable Text via Gradient-Operator Retrieval

Zizheng Zhang, Yuyang Liao, Chen Chen, Jian He, Dun Wu, Qianjin Yu, Yanqin Gao, Jin Yang, Kailai Zhang, Eng Siong Chng, Xionghu Zhong · 2026

Read TextBFGS: Quasi-Newton Optimization for Discrete Executable Text via Gradient-Operator Retrieval by Zizheng Zhang, Yuyang Liao, Chen Chen, Jian He, Dun Wu, Qianjin Yu, Yanqin Gao, Jin Yang, Kailai Zhang, Eng Siong Chng, Xionghu Zhong (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-2602-00059.txt.

Description

Zizheng Zhang and coauthors treat language model code repair as search guided by text feedback, and they note that stateless self correction wastes past lessons. Their system, TextBFGS, keeps a case base of past error to operator paths, much as a quasi Newton method keeps a sense of curvature. It reuses those patterns for new errors. Given a fresh error, it pulls like cases, applies their abstract fixes, and stores wins back into the base. On Python tasks from HumanEval and MBPP, it beats stateless baselines with fewer model calls.

Catalog record

CREATORZizheng Zhang, Yuyang Liao, Chen Chen, Jian He, Dun Wu, Qianjin Yu, Yanqin Gao, Jin Yang, Kailai Zhang, Eng Siong Chng, Xionghu Zhong
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE35732
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602.00059

Browse the full library · About the Tower of Babel