SOURCE-LINKED INTELLIGENCE

Large Language Models Develop Belief State Geometry In-Context

arXiv · AI, language, vision and robotics · article · Sep 15, 2026 · UTC

Large language models (LLMs) trained on next-token prediction exhibit remarkable in-context learning (ICL) abilities, yet the representations that support ICL remain poorly understood. We consider such representations in a controlled setting: prompting LLMs with data emitted from hidden Markov models (HMMs) and probing for the corresponding belief state -- the posterior distribution over the HMM's hidden states given the observed token history. Across six open-source LLMs prompted with data from 40 HMMs selected for non-trivial belief structure, we find that belief states are linearly decodabl

Read original source ↗ Open in workspace

recordType

paper

region

Global