预印本比较 AI 与人工审阅临床试验论文的查错表现Chinese only
一份预印本称，在植入错误的测试中，15 款 AI 模型的查错率为 53%—96%，未经训练的人工审稿者平均为 27%。在 58 篇 2026 年新发表的试验论文上，Claude Opus 5.5 得分最高，达 99%；这些数字仍是预印本报告的结果。
1 developments2 reports1 sources
Event developments 1
Preprint compares AI and human error detection in clinical trial papers
A preprint reports that 15 AI models caught 53%–96% of planted errors, versus an average of 27% for untrained human reviewers. On 58 newly published clinical trials from 2026, Claude Opus 5.5 scored highest at 99%. These are results reported in a preprint.
View source reports · 1 sources · 2 reports
- Rohan Paul (@rohanpaul_ai) · Preprint compares AI and human error detection in clinical trial papers
- Rohan Paul (@rohanpaul_ai) · Preprint test reports higher error detection by AI in clinical trial papers
All reports 2
Preprint test reports higher error detection by AI in clinical trial papers
In a preprint test using planted errors, 15 AI models detected 53% to 96% of errors, compared with an average of 27% for untrained human reviewers. The reported pattern also held across 58 new trials published in 2026; these figures come from discussion of a preprint and do not establish performance across all peer review settings.
1 more source reported this
- Rohan Paul (@rohanpaul_ai)
Preprint compares AI and human error detection in clinical trial papers
A preprint reports that 15 AI models caught 53%–96% of planted errors, versus an average of 27% for untrained human reviewers. On 58 newly published clinical trials from 2026, Claude Opus 5.5 scored highest at 99%. These are results reported in a preprint.
1 more source reported this
- Rohan Paul (@rohanpaul_ai)