模型即裁判论文趋势 2023–2026

主要结论

• 模型即裁判论文从 2023 年的 8 篇增长到 2026 年的 219 篇，三年间扩大到约 27 倍。
• NeurIPS 是发模型即裁判论文最多的会议，2023–2026 年共 173 篇，占四个会议总量的 43%。
• 交叉方向里涨得最快的是AI 智能体：同时属于两个主题的论文从 2023 年的 1 篇增至 2026 年的 78 篇。
• 在「评测与基准」的 5 个主题里，它的 2026 年论文数排第 2；同组增长最快的大模型基准是 154 倍，模型即裁判是 27 倍。

代表论文

按年份和会议均匀挑选，方便快速看到每一届的样貌

• NeurIPS 2026A Free Lunch in LLM Compression: Revisiting Retraining after Pruning
• ICML 2026A Judge-Aware Ranking Framework for Evaluating Large Language Models without Ground Truth
• ICLR 2026Addressing Pitfalls in the Evaluation of Uncertainty Estimation Methods for Natural Language Generation
• CVPR 2026Pixels Dont Lie But Your Detector Might Bootstrapping MLLM-as-a-Judge for
• NeurIPS 2025$\texttt{BetaConform}$: Efficient MAP Estimation of LLM Ensemble Judgment Performance with Prior Transfer
• ICML 2025Active Evaluation Acquisition for Efficient LLM Benchmarking
• ICLR 2025An Empirical Analysis of Uncertainty in Large Language Model Evaluations
• CVPR 2025ASHiTA: Automatic Scene-grounded HIerarchical Task Analysis
• NeurIPS 2024BertaQA: How Much Do Language Models Know About Local Culture?
• ICML 2024CHAI: Clustered Head Attention for Efficient LLM Inference
• ICLR 2024PandaLM: An Automatic Evaluation Benchmark for LLM Instruction Tuning Optimization
• NeurIPS 2023Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena
• ICML 2023SAM operates far from home: eigenvalue regularization as a dynamical phenomenon
• ICLR 2023On the complexity of nonsmooth automatic differentiation
• NeurIPS 2026A Stratified Multi-Rater Evaluation of LLM-Based Virtual Standardized Patients with a Deployed Data-Generation Platform
• ICML 2026Are LLM Evaluators Really Narcissists? Sanity Checking Self-Preference Evaluations
• ICLR 2026AutoMetrics: Approximate Human Judgments with Automatically Generated Evaluators
• CVPR 20263DrawAgent Teaching LLM to Draw in 3D with Early Contrastive
• NeurIPS 2025Adaptive Prediction-Powered AutoEval with Reliability and Efficiency Guarantees
• ICML 2025Evaluating Judges as Evaluators: The JETTS Benchmark of LLM-as-Judges as Test-Time Scaling Evaluators

全部模型即裁判论文

共 112 篇，第 2 / 3 页

51. ICLR 2026SocialHarmBench: Revealing LLM Vulnerabilities to Socially Harmful Requests
52. ICLR 2026SparseEval: Efficient Evaluation of Large Language Models by Sparse Optimization
53. ICLR 2026STAR: Strategy-driven Automatic Jailbreak Red-teaming For Large Language Model
54. ICLR 2026Talk, Evaluate, Diagnose: User-aware Agent Evaluation with Automated Error Analysis
55. ICLR 2026The Ideation-Execution Gap: Execution Outcomes of LLM-Generated versus Human Research Ideas
56. ICLR 2026The Unseen Frontier: Pushing the Limits of LLM Sparsity with Surrogate-Free ADMM
57. ICLR 2026Towards Reliable Benchmarking: A Contamination Free, Controllable Evaluation Framework for Multi-step LLM Function Calling
58. ICLR 2026TrustJudge: Inconsistencies of LLM-as-a-Judge and How to Alleviate Them
59. ICLR 2026Unlocking Full Efficiency of Token Filtering in Large Language Model Training
60. ICLR 2026Watermarking Diffusion Language Models
61. ICLR 2026WebDevJudge: Evaluating (M)LLMs as Critiques for Web Development Quality
62. ICLR 2026Your Agent May Misevolve: Emergent Risks in Self-evolving LLM Agents
63. ICLR 2025A Statistical Framework for Ranking LLM-based Chatbots
64. ICLR 2025AgentSquare: Automatic LLM Agent Search in Modular Design Space
65. ICLR 2025An Empirical Analysis of Uncertainty in Large Language Model Evaluations
66. ICLR 2025BadJudge: Backdoor Vulnerabilities of LLM-As-A-Judge
67. ICLR 2025Better than Your Teacher: LLM Agents that learn from Privileged AI Feedback
68. ICLR 2025Beyond correlation: The impact of human uncertainty in measuring the effectiveness of automatic evaluation and LLM-as-a-judge
69. ICLR 2025Catastrophic Failure of LLM Unlearning via Quantization
70. ICLR 2025Cheating Automatic LLM Benchmarks: Null Models Achieve High Win Rates
71. ICLR 2025Combatting Dimensional Collapse in LLM Pre-Training Data via Submodular File Selection
72. ICLR 2025Controlled LLM Decoding via Discrete Auto-regressive Biasing
73. ICLR 2025Diversity Empowers Intelligence: Integrating Expertise of Software Engineering Agents
74. ICLR 2025Do LLMs have Consistent Values?
75. ICLR 2025GraphEval: A Lightweight Graph-Based LLM Framework for Idea Evaluation
76. ICLR 2025HShare: Fast LLM Decoding by Hierarchical Key-Value Sharing
77. ICLR 2025Judge Decoding: Faster Speculative Sampling Requires Going Beyond Model Alignment
78. ICLR 2025JudgeBench: A Benchmark for Evaluating LLM-Based Judges
79. ICLR 2025Justice or Prejudice? Quantifying Biases in LLM-as-a-Judge
80. ICLR 2025Language Model Alignment in Multilingual Trolley Problems
81. ICLR 2025Language Models Are Implicitly Continuous
82. ICLR 2025Latent Space Chain-of-Embedding Enables Output-free LLM Self-Evaluation
83. ICLR 2025Learning LLM-as-a-Judge for Preference Alignment
84. ICLR 2025Limits to scalable evaluation at the frontier: LLM as judge won’t beat twice the data
85. ICLR 2025LiveBench: A Challenging, Contamination-Limited LLM Benchmark
86. ICLR 2025On Calibration of LLM-based Guard Models for Reliable Content Moderation
87. ICLR 2025Pangea: A Fully Open Multilingual Multimodal LLM for 39 Languages
88. ICLR 2025Perplexed by Perplexity: Perplexity-Based Data Pruning With Small Reference Models
89. ICLR 2025Progressive Mixed-Precision Decoding for Efficient LLM Inference
90. ICLR 2025RevisEval: Improving LLM-as-a-Judge via Response-Adapted References
91. ICLR 2025RMB: Comprehensively benchmarking reward models in LLM alignment
92. ICLR 2025RocketEval: Efficient automated LLM evaluation via grading checklist
93. ICLR 2025SeedLM: Compressing LLM Weights into Seeds of Pseudo-Random Generators
94. ICLR 2025Selective Attention Improves Transformer
95. ICLR 2025SpinQuant: LLM Quantization with Learned Rotations
96. ICLR 2025SqueezeAttention: 2D Management of KV-Cache in LLM Inference via Layer-wise Optimal Budget
97. ICLR 2025Style Outweighs Substance: Failure Modes of LLM Judges in Alignment Benchmarking
98. ICLR 2025ToolACE: Winning the Points of LLM Function Calling
99. ICLR 2025Trust or Escalate: LLM Judges with Provable Guarantees for Human Agreement
100. ICLR 2024An LLM can Fool Itself: A Prompt-Based Adversarial Attack

常见问题

2023–2026 年四大会议一共有多少篇模型即裁判论文？

共 401 篇，按年份分别是 2023 年 8 篇、2024 年 48 篇、2025 年 126 篇、2026 年 219 篇。

NeurIPS 2026 有多少篇模型即裁判论文？

NeurIPS 2026 有 88 篇模型即裁判论文，上一届是 53 篇。

ICLR 2026 有多少篇模型即裁判论文？

ICLR 2026 有 62 篇模型即裁判论文，上一届是 37 篇。

ICML 2026 有多少篇模型即裁判论文？

ICML 2026 有 64 篇模型即裁判论文，上一届是 33 篇。

CVPR 2026 有多少篇模型即裁判论文？

CVPR 2026 有 5 篇模型即裁判论文，上一届是 3 篇。

哪个会议的模型即裁判论文最多？

NeurIPS，2023–2026 年共 173 篇。