Computer Science · Artificial Intelligence
A preprint study found that the same hidden instruction could produce very different results in AI agents depending on whether it arrived as tool information or tool output, and what surrounded it. The findings have not been peer reviewed and may change.
This review-style evaluation compares two large language models—OpenAI o1 and DeepSeek R1—in English and Chinese for supporting junior anesthesiologists. Using 30 Delphi-created anesthesia crisis scenarios, experts rated accuracy and logic, while junior physicians rated clarity and usefulness. Results differed: OpenAI o1 scored higher for accuracy, while the DeepSeek R1 Chinese version scored higher for practical, step-by-step guidance.
