MRKL Systems: A Modular, Neuro-Symbolic Architecture That Combines Large Language Models, External Knowledge Sources and Discrete Reasoning

LLM + external tools/modules

Foundational work on LLM-based agents and LLM + external tools/modules.

Addresses how language models can extend beyond single-turn generation through LLM + external tools/modules.

The paper operationalizes LLM + external tools/modules through a structured agent mechanism.

The system connects an LLM with state, control, tools, memory, environments, interfaces, or other agents.

Uses task-specific datasets, environments, benchmarks, or interaction traces appropriate to the research question.

Uses the training or inference adaptation documented by the primary source; undisclosed details are not inferred.

Evaluates the proposed mechanism against relevant baselines in the target setting.

Compares against the underlying model, simpler control strategies, or prior agent mechanisms.

Reports evidence for the proposed contribution within its stated benchmark or environment.

Contributes a reusable pattern for building or evaluating LLM-based agents.

Primary source: https://arxiv.org/abs/2205.00445