FLEx: Language Modeling with F ew-shot L anguage Ex planations
Adar Avsian, Christopher Richardson, Anirudh Sundar, Larry Heck · 2026
Read FLEx: Language Modeling with F ew-shot L anguage Ex planations by Adar Avsian, Christopher Richardson, Anirudh Sundar, Larry Heck (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-2601-04157.txt.
Description
Adar Avsian, Christopher Richardson, Anirudh Sundar, and Larry Heck improve language models with just a few written explanations of past mistakes. Models err, and the same errors recur on related questions. Expert explanations can fix the pattern, but collecting many is costly where experts are scarce.
Their method, FLEx, clusters model errors by embedding to pick representative cases, checks that each paired explanation truly repairs its error, then sums the lessons into a short prompt prefix. That prefix is added at run time, with no change to model weights. On CounterBench, GSM8K, and ReasonIF, FLEx beats chain-of-thought prompting throughout. The four authors report it removes a large share of the errors that chain-of-thought leaves behind, up to the great majority on the sets tried.
Catalog record
| CREATOR | Adar Avsian, Christopher Richardson, Anirudh Sundar, Larry Heck |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 60 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2601-04157