Large Language Models (LLMs) often struggle with complex reasoning and planning tasks.

The TMK Framework

A new study published on arXiv explores the use of the Task-Method-Knowledge (TMK) framework to improve the reasoning abilities of LLMs. TMK, already known in the field of cognitive and educational science, stands out for its ability to capture causal, teleological, and hierarchical reasoning structures.

Study Details

The research evaluates TMK using the PlanBench benchmark, focusing on the Blocksworld domain. The goal is to verify whether TMK-structured prompting can help language models decompose complex planning problems into more manageable sub-tasks. The results highlight a significant improvement in the performance of reasoning models.

Results

TMK prompting enabled the reasoning model to achieve an accuracy of 97.3% on opaque, symbolic tasks (Random versions of Blocksworld in PlanBench), where it previously failed (31.5%). This suggests that TMK functions not merely as context, but also as a mechanism that steers reasoning models to activate formal code execution pathways.

For those evaluating on-premise deployments, there are trade-offs to consider. AI-RADAR offers analytical frameworks on /llm-onpremise to evaluate these aspects.