Task-Level Natural Language Priors as Learning Signals for Low-Resource LLM Training.

arXiv:2609.02244v2 Announce Type: replace Abstract: Large language models (LLMs) often struggle when low-resource training data are ambiguous or incomplete. Task-level natural-language priors can provide useful guidance in such settings, but existing approache