A new study analyzing 2,000 academic papers has found that large language models like GPT and Claude tend to focus their identified weaknesses on a paper's appendix, drawing over 50% of their criticisms from this section. This contrasts sharply with human reviewers, who extract only 8.4% of their critical feedback from appendices and focus more on the paper's core ideas and presentation. The LLMs' criticisms often delve into fine-grained technical details within the appendix that are not central to the paper's main contributions. AI
IMPACT Highlights a potential bias in AI-assisted academic reviewing, suggesting a need for better alignment with human review priorities.
RANK_REASON The cluster contains an academic paper analyzing the behavior of AI models.
Read on Hugging Face Daily Papers →
- arXiv
- Association for Computational Linguistics
- .claude
- Colm
- Empirical Methods in Natural Language Processing
- generative pre-trained transformer
- Hugging Face
- International Conference on Learning Representations
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →