●builderYou can significantly improve data extraction reliability by adding explicit constraints against guessing in your prompts.
●researcherThis highlights the sensitivity of LLM calibration to specific negative constraints in zero-shot extraction tasks.