WMF-AM: Probing LLM Working Memory via Depth-Parameterized Cumulative State Tracking

arXiv:2603.27343v2 Announce Type: replace Abstract: Existing large language models (LLMs) evaluations use fixed-difficulty benchmarks that cannot adapt as models improve, and rarely isolate specific cognitive processes. We introduce Working Memory Fidelity-Active Manipulation (WMF-AM), a probe of...