Red Teaming for Large Language Models At Scale: Tackling Hallucinations on Mathematics Tasks
This addresses the reliability of LLMs for mathematical reasoning, though it is incremental as it builds on existing red teaming methods.
The researchers tackled the problem of evaluating how red teaming techniques affect large language models' performance on mathematical tasks, finding that while structured reasoning and examples slow quality deterioration, GPT-3.5 and GPT-4 remain poorly suited for elementary calculations even with red teaming.
We consider the problem of red teaming LLMs on elementary calculations and algebraic tasks to evaluate how various prompting techniques affect the quality of outputs. We present a framework to procedurally generate numerical questions and puzzles, and compare the results with and without the application of several red teaming techniques. Our findings suggest that even though structured reasoning and providing worked-out examples slow down the deterioration of the quality of answers, the gpt-3.5-turbo and gpt-4 models are not well suited for elementary calculations and reasoning tasks, also when being red teamed.