Conjectured reduction of reasoning gap via specialized prompting

Establish whether specialized prompting strategies—including chain-of-thought, tree-of-thought, chain-of-code, chain-of-hindsight, program-of-thought, algorithmic skill prompting, and progressive hint prompting—reduce the reasoning gap for large language models on functional benchmarks.

Background

The paper surveys prompting methods that elicit explicit reasoning steps and notes their demonstrated accuracy benefits. The authors conjecture that these methods would reduce the reasoning gap, inviting empirical or theoretical validation on functionalized benchmarks.