EMNLP survey: 24% of authors say AI reviews worse than human, praise seen as fine
m2saxon · x · 2026-10-25
m2saxon shares a slide from EMNLP 2026 on author feedback about AI-generated peer reviews, quipping that authors broadly consider AI praise of their papers high quality while dismissing its criticisms.
The slide shows main complaints: 24% (68) rated AI reviews worse than human reviews, 17% (49) found them superficial, and 15% (42) said they missed the paper's main insight. Yet some authors noted the AI comments overlapped significantly with their human reviews and mirrored their limitations sections, while others said the AI felt forced to invent weaknesses, requested experiments already in the appendix, or conflicted with ARR guidelines.
More from Research
- Thread Details: Continual Training Across 11 Deepfake Generators, 4 Fairness Metrics — mundt_martin · 2026-10-25
- Continual Learning Keeps Deepfake Detectors Accurate but Worsens Fairness, Study Finds — mundt_martin · 2026-10-25
- Protenix removes the AlphaFold3 gatekeeper for labs worldwide — RichardsonDx · 2026-10-25
- ByteDance's open-source Protenix overtakes AlphaFold3 across FoldBench benchmarks — RichardsonDx · 2026-10-25
- How small can an LLM be and still speak coherent English? Sub-10M models fail — Holiday-Inspection75 · 2026-10-25
- EMNLP 2026 papers show LLM rankings shift with user knowledge level in multi-turn evals — IanArawjo · 2026-10-25