Reproduce “When Form Changes but Logic Doesn't: Building Logic-invariant LLMs through Logical Structure” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Xuyuan Liu, Xinshuai Dong, Elynn Chen, Yujun Yan

The authors introduce LoGIST, which represents formal rules as reduced ordered binary decision diagrams and encodes their structure as embeddings that language models consult through cross-attention. On tested synthetic reasoning tasks, LoGIST reduces inconsistency across equivalent rule rewrites; in one GRPO setup, Mixed inconsistency falls from 37.1% to 16.6%. These findings concern the formal-reasoning settings evaluated.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.