← Search

Shijie Lu

2 accepted papers

2025

QA‐LIGN: Aligning LLMs through Constitutionally Decomposed QA

EMNLP 2025

Alignment of large language models (LLMs) with principles like helpfulness, honesty, and harmlessness typically relies on scalar rewards that obscure which objectives drive the training signal. We introduce QA-LIGN, which decomposes monolithic rewards into interpretable principle-specific evaluation

Cited by 0SourcePDFScholar
2025

ToW: Thoughts of Words Improve Reasoning in Large Language Models

NAACL 2025long

We introduce thoughts of words (ToW), a novel training-time data-augmentation method for next-word prediction. ToW views next-word prediction as a core reasoning task and injects fine-grained thoughts explaining what the next word should be and how it is related to the previous contexts in pre-train…