对齐与安全论文趋势 2023–2026

RLHF、偏好优化、红队与安全对齐

统计范围：NeurIPS、ICLR、ICML、CVPR 四个会议，2023–2026 年，共 4,558 篇被归入该主题的论文。

主要结论

每年论文数

各会议论文数

按年份和会议统计

对齐与安全论文数，按年份和会议
年份NeurIPSICLRICMLCVPR合计
202313887101112438
2024348156246125875
20254554733001351,363
20264806235512281,882

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

全部对齐与安全论文

年份全部2023202420252026
会议全部NeurIPSICLRICMLCVPR

共 473 篇，第 6 / 10 页

  1. ICLR 2025LLaMaFlex: Many-in-one LLMs via Generalized Pruning and Weight Sharing
  2. ICLR 2025LLaVA-MoD: Making LLaVA Tiny via MoE-Knowledge Distillation
  3. ICLR 2025LLM Unlearning via Loss Adjustment with Only Forget Data
  4. ICLR 2025Local Patterns Generalize Better for Novel Anomalies
  5. ICLR 2025Locality Alignment Improves Vision-Language Models
  6. ICLR 2025Logicbreaks: A Framework for Understanding Subversion of Rule-based Inference
  7. ICLR 2025Long-tailed Adversarial Training with Self-Distillation
  8. ICLR 2025LongPO: Long Context Self-Evolution of Large Language Models through Short-to-Long Preference Optimization
  9. ICLR 2025Looking Inward: Language Models Can Learn About Themselves by Introspection
  10. ICLR 2025MA-RLHF: Reinforcement Learning from Human Feedback with Macro Actions
  11. ICLR 2025Machine Unlearning Fails to Remove Data Poisoning Attacks
  12. ICLR 2025MACPO: Weak-to-Strong Alignment via Multi-Agent Contrastive Preference Optimization
  13. ICLR 2025Magnetic Preference Optimization: Achieving Last-iterate Convergence for Language Model Alignment
  14. ICLR 2025Magpie: Alignment Data Synthesis from Scratch by Prompting Aligned LLMs with Nothing
  15. ICLR 2025MallowsPO: Fine-Tune Your LLM with Preference Dispersions
  16. ICLR 2025MAP: Multi-Human-Value Alignment Palette
  17. ICLR 2025Mask-DPO: Generalizable Fine-grained Factuality Alignment of LLMs
  18. ICLR 2025Maximizing the Potential of Synthetic Data: Insights from Random Matrix Theory
  19. ICLR 2025Measuring memorization in RLHF for code completion
  20. ICLR 2025Measuring Non-Adversarial Reproduction of Training Data in Large Language Models
  21. ICLR 2025MIA-Bench: Towards Better Instruction Following Evaluation of Multimodal LLMs
  22. ICLR 2025MIA-DPO: Multi-Image Augmented Direct Preference Optimization For Large Vision-Language Models
  23. ICLR 2025Mind Control through Causal Inference: Predicting Clean Images from Poisoned Data
  24. ICLR 2025Mitigate the Gap: Improving Cross-Modal Alignment in CLIP
  25. ICLR 2025Mitigating Memorization in Language Models
  26. ICLR 2025Mitigating Reward Over-Optimization in RLHF via Behavior-Supported Regularization
  27. ICLR 2025Mitigating Spurious Correlations in Zero-Shot Multimodal Models
  28. ICLR 2025Mitigating the Backdoor Effect for Multi-Task Model Merging via Safety-Aware Subspace
  29. ICLR 2025Mix-CPT: A Domain Adaptation Framework via Decoupling Knowledge Learning and Format Alignment
  30. ICLR 2025Mix-LN: Unleashing the Power of Deeper Layers by Combining Pre-LN and Post-LN
  31. ICLR 2025MMDT: Decoding the Trustworthiness and Safety of Multimodal Foundation Models
  32. ICLR 2025Model Editing as a Robust and Denoised variant of DPO: A Case Study on Toxicity
  33. ICLR 2025Modeling Future Conversation Turns to Teach LLMs to Ask Clarifying Questions
  34. ICLR 2025Monitoring Latent World States in Language Models with Propositional Probes
  35. ICLR 2025Montessori-Instruct: Generate Influential Training Data Tailored for Student Learning
  36. ICLR 2025Moral Alignment for LLM Agents
  37. ICLR 2025More RLHF, More Trust? On The Impact of Preference Alignment On Trustworthiness
  38. ICLR 2025Multi-objective antibody design with constrained preference optimization
  39. ICLR 2025Navigating Neural Space: Revisiting Concept Activation Vectors to Overcome Directional Divergence
  40. ICLR 2025Near-Exact Privacy Amplification for Matrix Mechanisms
  41. ICLR 2025Neural Interactive Proofs
  42. ICLR 2025NNsight and NDIF: Democratizing Access to Open-Weight Foundation Model Internals
  43. ICLR 2025No Preference Left Behind: Group Distributional Preference Optimization
  44. ICLR 2025OASIS Uncovers: High-Quality T2I Models, Same Old Stereotypes
  45. ICLR 2025On a Connection Between Imitation Learning and RLHF
  46. ICLR 2025On Calibration of LLM-based Guard Models for Reliable Content Moderation
  47. ICLR 2025On Evaluating the Durability of Safeguards for Open-Weight LLMs
  48. ICLR 2025On Large Language Model Continual Unlearning
  49. ICLR 2025On Linear Representations and Pretraining Data Frequency in Language Models
  50. ICLR 2025On Targeted Manipulation and Deception when Optimizing LLMs for User Feedback

在 Claude、Cursor 里直接问这个趋势

接上 Paperlist MCP 后，写综述、挑方向或准备投稿时，不用离开对话框就能拿到这页的数字：每年各会议的论文数、增长倍数、涨得最快的子方向和代表论文，AI 会按你的问题整理。

可以直接复制这句话问 AI

用 paperlist 查对齐与安全在 NeurIPS 2023–2026 的论文趋势，哪一年涨得最快？

AI 会调用的工具：get_topic_stats {"topic": "alignment-and-safety"}

两分钟接入 Paperlist MCP（Claude Code、Cursor、Codex、ChatGPT）

常见问题

2023–2026 年四大会议一共有多少篇对齐与安全论文？
共 4,558 篇，按年份分别是 2023 年 438 篇、2024 年 875 篇、2025 年 1,363 篇、2026 年 1,882 篇。
NeurIPS 2026 有多少篇对齐与安全论文？
NeurIPS 2026 有 480 篇对齐与安全论文，上一届是 455 篇。
ICLR 2026 有多少篇对齐与安全论文？
ICLR 2026 有 623 篇对齐与安全论文，上一届是 473 篇。
ICML 2026 有多少篇对齐与安全论文？
ICML 2026 有 551 篇对齐与安全论文，上一届是 300 篇。
CVPR 2026 有多少篇对齐与安全论文？
CVPR 2026 有 228 篇对齐与安全论文，上一届是 135 篇。
哪个会议的对齐与安全论文最多？
NeurIPS，2023–2026 年共 1,421 篇。

「对齐与负责任 AI」下的其他主题