CarO : C hain-of- A nalogy R easoning O ptimization for Robust Content Moderation
Bingzhe Wu [email protected], Haotian Lu [email protected], Yuchen Mou [email protected] · 2026
Read CarO : C hain-of- A nalogy R easoning O ptimization for Robust Content Moderation by Bingzhe Wu [email protected], Haotian Lu [email protected], Yuchen Mou [email protected] (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-2604-10504.txt.
Description
Wu, Lu, and Mou harden moderation models against sly context traps. Even reasoning tuned models trip on vague cases when context plants false shortcuts. Their CARO frame first grows analogy chains from moderation data with retrieval plus tuning. A second preference pass then rewards analogy habits outright. Tests beat top reasoning and guard models, lifting mean F1 by 24.9 percent on hard vague sets.
Catalog record
| CREATOR | Bingzhe Wu [email protected], Haotian Lu [email protected], Yuchen Mou [email protected] |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 40442 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.10504