2025年的一项研究显示,在评估科学说法时ChatGPT经常会给出不准确和不一致的答案,其识别虚假陈述的准确率只有16.4%. A 2025 study reveals ChatGPT often gives inaccurate, inconsistent answers when assessing scientific claims, with only 16.4% accuracy in identifying false statements.

华盛顿州立大学的研究人员在2025年的一项研究发现,当评估科学主张时,ChatGPT经常会给出不准确和不一致的答案,其精度仅比随机猜测优于60%. A 2025 study by Washington State University researchers finds ChatGPT frequently gives inaccurate and inconsistent answers when evaluating scientific claims, with accuracy only 60% better than random guessing. 在700个科学陈述中进行了10次重复测试,AI正确地识别出错误的说法只有16.4%时间,并且只提供73%时间的一致反应. Tested on 700 scientific statements repeated 10 times, the AI correctly identified false claims just 16.4% of the time and provided consistent responses only 73% of the time. 尽管准确性从2024年的75%略有提高到2025年的80%,但研究人员警告用户 - - 特别是专业人士 - - 要验证人工智能输出并保持怀疑,因为该模型缺乏真正的推理并且依赖于模式记忆. Despite a slight accuracy improvement from 75% in 2024 to 80% in 2025, researchers warn users—especially professionals—to verify AI outputs and remain skeptical, as the model lacks true reasoning and relies on pattern memorization.