Language Model Planning from an Information Theoretic Perspective
This work addresses an open question in AI interpretability and model design, providing insights into LM planning mechanisms, though it is incremental in advancing analysis methods.
The researchers tackled the problem of understanding how decoder-only language models engage in planning for long-range generation by analyzing hidden states using a compression and mutual information framework, revealing task-dependent planning horizons, implicit preservation of alternative continuations, and reliance on recent computations.
The extent to which decoder-only language models (LMs) engage in planning, that is, organizing intermediate computations to support coherent long-range generation, remains an open and important question, with implications for interpretability, reliability, and principled model design. Planning involves structuring computations over long horizons, considering multiple possible continuations, and selectively reusing past information, but how effectively transformer-based LMs realize these capabilities is still unclear. We address these questions by analyzing the hidden states at the core of transformer computations, which capture intermediate results and act as carriers of information. Since these hidden representations are often redundant and encumbered with fine-grained details, we develop a pipeline based on vector-quantized variational autoencoders that compresses them into compact summary codes. These codes enable measuring mutual information, allowing systematic analysis of the computational structure underlying model behavior. Using this framework, we study planning in LMs across synthetic grammar, path-finding tasks, and natural language datasets, focusing on three key aspects: (i) the planning horizon of pre-output computations, (ii) the extent to which the model considers alternative valid continuations, and (iii) the reliance of new predictions on earlier computations. By answering these questions, we advance the understanding of how planning is realized in LMs and contribute a general-purpose pipeline for probing the internal dynamics of LMs and deep learning systems. Our results reveal that the effective planning horizon is task-dependent, that models implicitly preserve information about unused correct continuations, and that predictions draw most on recent computations, though earlier blocks remain informative.