AI Paper Atlas

Research map / Large language models / Alignment and preferences

References Improve LLM Alignment in Non-Verifiable Domains

Kejian Shi, Yixin Liu, PeiFeng Wang, Alexander Fabbri, Shafiq Joty, Arman Cohan

ICLR 2026 · Large language models › Alignment and preferences · approach: Steering and correcting LLM behavior · 2 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add reference output with explicit usage instructions (RefEval vs reference-free LLMBar-Base)LLM judge · 5 pairwise datasets (11 judges)accuracy (%): 72.3 -> 79.1Improved
add references with RefEval prompt over reference-free Ref-Free promptLLM judge · 5 pairwise datasets (11 judges)accuracy (%): 73.7 -> 79.1Improved
naive reference inclusion (LLMBar-Ref) over LLMBar-BaseLLM judge · 5 pairwise datasets (11 judges)accuracy (%): 72.3 -> 74.0Improved
add RefEval to small judge llama-3-8bllama-3-8b judge · 5 pairwise datasetsaccuracy (%): 60.1 -> 77.5Improved
add RefEval to strong judge qwen-2.5-72bqwen-2.5-72b judge · 5 pairwise datasetsaccuracy (%): 79.4 -> 84.6Improved
use human-edited Oracle reference with RefEval (GPT-4o judge)GPT-4o judge · LLMBar-Adversarialaccuracy (%): 86.8 -> 88.4Improved
use weaker references (Mistral-Nemo-12B) with RefEval vs reference-free (ablation)Llama-3.1-70B judge · 4 pairwise datasetsaccuracy (%): 80.3 -> 85.0Improved
add multi-reference voting to RefEvalRefEval · 4 datasets (11 judges)accuracy: 0.782 -> 0.789Improved
add DPO stage with reference-free self-judge after distillationLlama-3-8B-Instruct · AlpacaEval LCwin rate (%): 53.9 -> 67.5Improved
use reference-guided self-judge instead of reference-free self-judge in DPOLlama-3-8B-Instruct · AlpacaEval LCwin rate (%): 67.5 -> 73.1Improved
use reference-guided self-judge instead of reference-free self-judge in DPOQwen2.5-7B-SFT · Arena-Hardscore (%): 71.8 -> 74.1Improved
use RefEval judge instead of ROUGE reward in DPOLlama-3-8B-Instruct · AlpacaEval LCwin rate (%): 56.4 -> 73.1Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy