Not All Queries Need Deep Thought: CoFiCot for Adaptive Coarse-to-fine Stateful Refinement

2026-03-09

AI Summary

This work proposes CoFiCot, a novel framework addressing the inefficiency of uniform computational resource allocation in large language models during reasoning, which often leads to over-correction on simple tasks and under-correction on complex ones. CoFiCot introduces a multi-metric difficulty classifier that dynamically assesses problem complexity by integrating semantic entropy, consensus reliability, and predictive inference. Based on this assessment, it applies differentiated refinement strategies: efficient aggregation for simple queries and a context-aware, stateful sequential correction loop for complex ones. Coupled with a process reward model, this approach balances fine-grained error localization with global logical consistency, effectively mitigating resource waste and performance bottlenecks while significantly enhancing both reasoning quality and efficiency across tasks of varying difficulty.

Technology Category

• Knowledge Representation and Reasoning: Computational Complexity of Reasoning
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning
• Search and Optimization: Metareasoning and Metaheuristics

Application Category

• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Scaling test-time computation enhances LLM reasoning ability but faces a uniform computation paradox. Allocating identical resources leads to over-correction on simple tasks and insufficient refinement on complex ones. To address this, we propose CoFiCot, a coarse-to-fine adaptive framework that dynamically tailors inference strategies to problem difficulty. Specifically, we implement a multi-metric classifier that triages queries by synthesizing semantic entropy, consensus reliability, and predicted reasoning depth . This enables a differentiated refinement stage that applies efficient aggregation for simple queries while routing complex ones to a context-aware correction loop . We formalize correction as a stateful sequential propagation process , where each repair is strictly conditioned on the verified history of prior rectifications. By integrating Process Reward Models (PRMs) within this state-dependent trajectory, CoFiCot effectively bridges the gap between granular error localization and global logical coherence, preventing the context fragmentation typical of stateless refinement methods.

Problem

• test-time computation
• uniform computation paradox
• adaptive refinement
• reasoning ability
• stateful correction

Innovation

• adaptive refinement
• stateful reasoning
• Process Reward Models
• semantic entropy
• coarse-to-fine inference