R0040/2026-04-01/Q002/SRC03

Research	R0040 — RLHF Alternatives
Run	2026-04-01
Query	Q002
Search	S02
Result	S02-R01
Source	SRC03

Fu et al. -- Reward Shaping to Mitigate Reward Hacking in RLHF

Source

Field	Value
Title	Reward Shaping to Mitigate Reward Hacking in RLHF
Publisher	arXiv
Author(s)	Jiayi Fu, Xuandong Zhao, Chengyuan Yao, Heng Wang, Qi Han, Yanghua Xiao
Date	2025-02-26 (revised 2026-01-21)
URL	https://arxiv.org/abs/2502.18770
Type	Research paper

Summary

Dimension	Rating
Reliability	Medium-High
Relevance	High
Bias: Missing data	Low risk
Bias: Measurement	Low risk
Bias: Selective reporting	Low risk
Bias: Randomization	N/A -- not an RCT
Bias: Protocol deviation	N/A -- not an RCT
Bias: COI/Funding	Low risk

Rationale

Dimension	Rationale
Reliability	Recent paper with reproducible results. Code publicly available. Tested on standard benchmarks.
Relevance	Directly demonstrates that RLHF can be fixed from within through reward shaping.
Bias flags	Academic paper without obvious commercial interest.

Evidence Extracts

Evidence ID	Summary
SRC03-E01	PAR method achieves 5+ point AlpacaEval win rate improvement while mitigating reward hacking