# RIMRULE: A New Approach for Adapting Language Models to Specific Tools RIMRULE is a new approach for adapting language models to specific tools, a common challenge in the field of artificial intelligence. Researchers have proposed a method called RIMRULE, which uses dynamic rule injection and an MDL objective to favor generality and conciseness of rules. The RIMRULE method is based on distilling compact and interpretable rules from failure traces of LLMs. These rules are then inserted into the prompt during inference to improve performance on specific tool activities. The method has been tested on tool-use benchmarks and shown improved accuracy compared to prompting-based adaptation methods. The portability of rules learned from one LLM has also been evaluated, with promising results. This suggests that the RIMRULE method may be applicable to a range of language models and specific tools without requiring reconfiguration.