🤖 AI Summary
This work proposes a training-free meta-reasoning framework that overcomes limitations of existing approaches—which often rely on backward rewards, coarse-grained actions, or auxiliary trained controllers—by leveraging a large language model (LLM) as a forward cognitive demand estimator. At each reasoning step, the framework dynamically assesses the remaining cognitive load required to solve a problem, guided by a 16-dimensional cognitive science scale encompassing facets such as attention, abstraction, and spatial reasoning for fine-grained diagnosis. Integrating residual demand signals, universal exemplars, and action intervention mechanisms, the method enables zero-shot transfer across diverse LLMs. Evaluated on three prominent LLMs and six reasoning benchmarks, it achieves an average accuracy improvement of 21.9% over direct prompting and 9% over standard chain-of-thought, with particularly pronounced gains in mathematical and programming tasks.
📝 Abstract
Recent meta-reasoning frameworks improve LLM reasoning by wrapping chain-of-thought generation in an iterative control loop, allowing more effective backtracking, termination of reasoning loops, and injection of promising reasoning patterns, among other strategy adjustments. Despite promising results, methods often rely on backward-looking reward functions, utilize coarse search actions, or require additional reasoning controller training requiring many-shot supervision. We introduce Cognitive Demand Steering (CDS), a training-free meta-reasoning framework equipped with residual demand assessment: at each step, an LLM-based progress evaluator characterizes the residual reasoning required to arrive at a solution rather than merely evaluating the previous step. This allows a meta-controller to select reasoning interventions comprising both general-purpose exemplars and actions (e.g., general guidance for quantitative reasoning) that directly tackle this forward-looking demand signal. This shift eliminates the need for any trained component while enabling zero-shot transfer across models and tasks with no adaptation. Rather than relying on coarse characterizations, we employ cognitive scales to both design interventions as well as profile initial problem complexity and residual demand signal over 16 dimensions motivated by cognitive science (e.g., attention and scan, learning and abstraction, spatio-physical reasoning), giving the controller a fine-grained vocabulary for diagnosing. Averaged across three frontier LLMs and six reasoning benchmarks, CDS improves accuracy by $21.9\%$ over direct calls and $9\%$ over standard CoT reasoning, with the largest gains on difficult mathematics and coding tasks.