评分 6 · 来源：cs.LG updates on arXiv.org · 发布于 2026-04-17
评分依据：Clean experimental methodology for studying LLM generalization using shortest path, good controlled setting
arXiv:2604.15306v1 Announce Type: cross Abstract: Whether language models can systematically generalize remains actively debated. Yet empirical performance is jointly shaped by multiple factors such as training data, training paradigms, and inference-time strategies, making failures difficult to interpret. We introduce a controlled synthetic environment based on shortest-path planning, a canonical composable sequential optimization problem.