Systematically demonstrates that 'easy-to-hard' curriculum learning provides no benefit for LLM deductive reasoning tasks.
March 31, 2026
Original Paper
Rethinking Easy-to-Hard: Limits of Curriculum Learning in Post-Training for Deductive Reasoning
arXiv · 2603.27226
The Takeaway
It challenges the deeply ingrained belief that sequencing training data by complexity helps models generalize better in logic or arithmetic. This result could significantly simplify training pipelines for reasoning-heavy post-training by proving random sampling is just as effective.
From the abstract
Curriculum learning (CL), motivated by the intuition that learning in increasing order of difficulty should ease generalization, is commonly adopted both in pre-training and post-training of large language models (LLMs). The intuition of CL is particularly compelling for compositional reasoning, where complex problems are built from elementary inference rules; however, the actual impact of CL on such tasks remains largely underexplored. We present a systematic empirical study of CL for post-trai