PRISM: A Dual View of LLM Reasoning through Semantic Flow and Latent Computation

2026-03-23

AI Summary

This work addresses the limitation of existing large language models (LLMs) in multi-step reasoning, which stems from the lack of joint analysis of semantic flow and internal hidden states, hindering precise failure diagnosis. The authors propose PRISM, a novel framework that, for the first time, simultaneously models cross-step semantic evolution and cross-layer hidden state dynamics to construct structured, observable representations of reasoning trajectories. By integrating probabilistic modeling, semantic trajectory analysis, and hidden state tracking, PRISM uncovers characteristic failure modes—such as verification loops, overthinking, and premature commitment—and elucidates how prompts modulate reasoning behavior. Experiments across multiple models and benchmarks demonstrate that PRISM enables fine-grained diagnostic insights into reasoning processes, surpassing conventional evaluation paradigms that rely solely on final accuracy.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Knowledge Representation and Reasoning: Diagnosis and Abductive Reasoning
• Natural Language Processing: (Large) Language Models

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Large language models for search

Abstract

Large language models (LLMs) solve complex problems by generating multi-step reasoning traces. Yet these traces are typically analyzed from only one of two perspectives: the sequence of tokens across different reasoning steps in the generated text, or the hidden-state vectors across model layers within one step. We introduce PRISM (Probabilistic Reasoning Inspection through Semantic and Implicit Modeling), a framework and diagnostic tool for jointly analyzing both levels, providing a unified view of how reasoning evolves across steps and layers. Across multiple reasoning models and benchmarks, PRISM uncovers systematic patterns in the reasoning process, showing that failed trajectories are more likely to become trapped in unproductive verification loops and further diverge into distinct modes such as overthinking and premature commitment, which behave differently once a candidate answer is reached. It further reveals how prompting reshapes reasoning behavior beyond aggregate accuracy by altering both semantic transitions and internal computational patterns. By modeling reasoning trajectories as structured processes, PRISM makes these behaviors observable and analyzable rather than relying solely on final-task accuracy. Taken together, these insights position PRISM as a practical tool for analyzing and diagnosing reasoning processes in LLMs.

Problem

• LLM reasoning
• reasoning trace analysis
• semantic flow
• latent computation
• reasoning failure

Innovation

• PRISM
• semantic flow
• latent computation
• reasoning trajectories
• probabilistic reasoning inspection