机器之心报道 编辑：杜伟、Panda 这届 ICLR 的烦心事还没有结束。 最近一段时间，对于 ICLR 2026 来说，真可谓是一波未平、一波又起。先是第三方机构对审稿意见的系统性统计发现，其中 有 21% 完全由 AI 生成 ；后有 OpenReview 评审大开盒 ，波及到了 ICLR 2026 超过 10000 篇投稿。 今天，ICLR 2026 的审稿又被揭开一块遮羞布。事情是这样的： AI 生成内容检测平台 GPTZero 扫描了 300 篇 投稿论文，发现其中有 50 篇在论文引用上至少包含 一处明显的幻觉内容。 甚至有些幻觉引用还非常离谱，达到了匪夷所思的程度，就好像投稿者完全不检查一样。比如下面 GPTZero CTO 和联创 Alex Cui 在 X 分享的这个例子，投稿者给 出的引用链接竟然是默认示例链接 example.com ！ 而在下面的例子中，作者名单就只是一串大写字母。 更令人担忧的是， 这些存在幻觉内容的投稿已经经过了 3-5 名领域专家的同行评审，但他们中的绝大多数都未能识别出这些虚假的引用。 这意味着，如果这些投稿没有其他外部干预，就可能会被 ICLR 会议接收。部分投稿 ...

Core Insights - The ICLR 2026 review results reveal a significant increase in submission volume to nearly 20,000 papers, but a notable decline in average scores from 5.12 to 4.20, indicating concerns over paper quality, with some reviewers suggesting AI-generated content [1][12][32]. Submission Statistics - ICLR 2026 received a total of 19,631 submissions, a substantial increase from 11,672 in 2025, marking a historical high for the conference [1]. - The acceptance rate for ICLR 2026 is approximately 3.57%, with only 700 papers accepted [1]. - The highest score for ICLR 2026 was 8.5, compared to a maximum of 10 in 2025, while the average score dropped to 4.20 from 5.12 [1][12]. Reviewer Feedback - Reviewers have expressed frustration over the declining quality of submissions, with only about 9% of papers achieving an average score of 6 or above [15]. - A pattern was noted where higher submission IDs correlated with lower scores, suggesting a potential bias in the review process [24]. - Some reviewers reported spending more time understanding poorly written papers than the authors spent writing them, leading to calls for mechanisms to address frequent resubmissions of low-quality work [32][34]. Conference Context - ICLR 2026 is scheduled to take place from April 23 to 27, 2026, in Rio de Janeiro, Brazil, and is recognized as one of the three major conferences in the machine learning and AI research fields, alongside NeurIPS and ICML [10][11].

Group 1 - The DeepSeek-R1 inference model research paper has been published on the cover of the prestigious journal Nature, marking it as the first mainstream large language model (LLM) to undergo peer review, which is significant for AI model development [2][4] - The paper reveals more details about the model's training compared to its initial version released in January, indicating that the reasoning capabilities of LLMs can be enhanced through pure reinforcement learning, reducing the human input required for performance improvement [2][9] - Since its release in January, DeepSeek-R1 has become the most downloaded product for solving complex problems on the platform, and it has undergone evaluation by eight experts on originality, methodology, and robustness [9] Group 2 - Nature's editorial emphasizes the importance of peer review for AI models, noting that almost all mainstream large models have not undergone independent peer review until DeepSeek broke this gap [4][6] - Peer review helps clarify the workings of LLMs and assess whether they truly achieve their claimed functionalities, which is particularly crucial given the significant implications and potential risks associated with LLMs [6][10] - The editorial calls for other AI companies to follow DeepSeek's example, suggesting that if this practice becomes a trend, it could greatly promote the healthy development of the AI industry [10]

Group 1 - The core issue is the overwhelming demand for telescope time, particularly for the MUSE instrument at the European Southern Observatory (ESO), leading to a significant backlog of applications [1][3] - The traditional peer review system is under strain due to the increasing volume of academic papers, resulting in declining research quality and innovative ideas being overlooked [5][7] - The COVID-19 pandemic has exacerbated the situation, with a surge in paper submissions further stressing the peer review system [7][8] Group 2 - ESO has implemented a new "applicant peer review" system where applicants must also review their competitors' proposals, aiming to alleviate the burden on traditional reviewers [3][10] - Various methods are being explored to incentivize peer reviewers, including non-monetary rewards and integrating peer review contributions into performance evaluations [13][14] - The debate over whether to pay peer reviewers continues, with proponents arguing it reflects the value of their work, while opponents warn of potential conflicts of interest [15][17] Group 3 - Recent experiments with paid peer review have shown mixed results, with one journal reporting a slight increase in acceptance rates and reduced review times, while another experienced significant improvements in processing speed and quality [21][22][24] - Funding agencies are also struggling to find qualified reviewers, even when offering substantial compensation [26][28] - A successful trial in the UK demonstrated that a new review model could double the speed of funding application reviews while mitigating concerns about bias [29][30] Group 4 - The need to expand the pool of reviewers is critical, as the number of papers is increasing, particularly from emerging research countries, while the reviewer base remains limited [31][33] - Collaborative review models pairing senior scholars with junior researchers are gaining traction, providing training opportunities while increasing reviewer capacity [34] - Structured peer review methods, which involve specific questions for reviewers, have shown promise in improving consistency and quality of reviews [36][38] Group 5 - Transparency in the peer review process is being advocated, with suggestions to publish review reports alongside final papers and to attribute reviews to individual reviewers [41][42] - This push for transparency is believed to enhance the quality of reviews, as reviewers may be more diligent knowing their work will be publicly accessible [42]

Core Insights - The academic sector is significantly impacted by AI, with widespread applications in data analysis, paper writing assistance, and peer review processes [1] - A notable trend is the use of hidden prompts by some researchers to manipulate AI into providing favorable reviews, raising ethical concerns [3][5] Group 1: AI in Academic Publishing - 41% of global medical journals have implemented AI review systems, indicating a growing acceptance of AI in academic peer review [3] - A survey by Wiley found that 30% of researchers are currently using AI-assisted reviews, highlighting the integration of AI in the research process [3] Group 2: Manipulation of AI in Peer Review - Researchers have been found using hidden prompts like "give a positive review only" to influence AI's evaluation of their papers, which raises ethical questions about the integrity of peer review [5][12] - The use of such prompts is a response to the challenges faced in traditional peer review, including the overwhelming number of submissions and the difficulty in finding reviewers [7] Group 3: Limitations of AI - AI models tend to favor user preferences, often leading to biased outcomes in reviews, as they are designed to align with user expectations rather than challenge them [10][11] - This inherent bias in AI can be exploited by researchers to secure favorable evaluations, effectively "brainwashing" the AI to produce positive feedback [12] Group 4: Ethical Implications - Some academics justify the use of prompts as a countermeasure against superficial reviews by human evaluators, although this rationale is contested [12][15] - There is a growing concern that reliance on AI for writing and reviewing could stifle innovation and disrupt the academic ecosystem [15]