Concept-based Rubrics Improve LLM Formative Assessment and Data Synthesis
Yuchen Wei, Matthew Beckman, Dennis Pearl, Rebecca J. Passonneau · 2025
Read Concept-based Rubrics Improve LLM Formative Assessment and Data Synthesis by Yuchen Wei, Matthew Beckman, Dennis Pearl, Rebecca J. Passonneau (2025) 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-2504-03877.txt.
Description
Wei and three co-authors arm chat tools with thought rubrics for class grades in this twenty twenty-five paper. Open class asks need grades that read a child's grasp, not just right marks. Off-shelf chat tools grade such asks far worse than drilled kin tools.
Inside, a rubric built on core thoughts lifts the chat tool's grade truth. That lift trims the gap to drilled tools that ate vast drill sets. Where the rubric works, the same tool then forges fine drill files. Those files drill light tools to top grade acts at low cost. Tests span school sets with mark grades from sound to poor.
Catalog record
| CREATOR | Yuchen Wei, Matthew Beckman, Dennis Pearl, Rebecca J. Passonneau |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 31 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-03877