AAAS Study: ChatGPT Mimics SciPak Briefs but Sacrifices Accuracy
A year-long AAAS experiment found ChatGPT can mimic SciPak brief structure but often sacrifices accuracy for simplicity—conflating correlation with causation, missing context, and overhyping results—leading journalists to rate LLM summaries poorly. The team concluded ChatGPT isn’t ready for SciPak briefs without extensive fact-checking; they may retest after major model updates.