NeurIPS 2025 accepted papers on Large language models
962 of the 5,812 papers accepted at NeurIPS 2025 (16.6%) are on Large language models, compared with 13.4% at NeurIPS 2024. They fall into 7 topics.
See NeurIPS 2025 in the interactive map
Working on something in this area? Describe your idea in scime atlas to see the closest accepted papers by meaning and how crowded the spot has become.
Topics and papers
Language, safety and interpretability · 255 papers
- Measuring what Matters: Construct Validity in Large Language Model Benchmarks (7 citations)
- TabDPT: Scaling Tabular Foundation Models on Real Data (6 citations)
- Artificial Hivemind: The Open-Ended Homogeneity of Language Models (and Beyond) (3 citations)
- MMAR: A Challenging Benchmark for Deep Reasoning in Speech, Audio, Music, and Their Mix (3 citations)
- SQLens: An End-to-End Framework for Error Detection and Correction in Text-to-SQL (3 citations)
- Why and How LLMs Hallucinate: Connecting the Dots with Subsequence Associations (3 citations)
- Audio Flamingo 3: Advancing Audio Intelligence with Fully Open Large Audio Language Models (2 citations)
- Concept-Guided Interpretability via Neural Chunking (2 citations)
- Efficient Pre-Training of LLMs via Topology-Aware Communication Alignment on More Than 9600 GPUs (2 citations)
- Fairshare Data Pricing via Data Valuation for Large Language Models (2 citations)
- Fine-grained List-wise Alignment for Generative Medication Recommendation (2 citations)
- Geometry of Decision Making in Language Models (2 citations)
and 243 more in the interactive map.
Reasoning and chain of thought · 225 papers
- The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity (29 citations)
- DAPO: An Open-Source LLM Reinforcement Learning System at Scale (19 citations)
- ReSearch: Learning to Reason with Search for LLMs via Reinforcement Learning (10 citations)
- T2I-R1: Reinforcing Image Generation with Collaborative Semantic-level and Token-level CoT (9 citations)
- What is Your Data Worth to GPT? LLM-Scale Data Valuation with Influence Functions (8 citations)
- VL-Rethinker: Incentivizing Self-Reflection of Vision-Language Models with Reinforcement Learning (7 citations)
- EAGLE-3: Scaling up Inference Acceleration of Large Language Models via Training-Time Test (6 citations)
- Delving into RL for Image Generation with CoT: A Study on DPO vs. GRPO (5 citations)
- Latent Chain-of-Thought for Visual Reasoning (5 citations)
- Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach (5 citations)
- CAD-Coder: Text-to-CAD Generation with Chain-of-Thought and Geometric Reward (4 citations)
- KVLink: Accelerating Large Language Models via Efficient KV Cache Reuse (4 citations)
and 213 more in the interactive map.
Agents, code and math · 198 papers
- PSI: A Benchmark for Human Interpretation and Response in Traffic Interactions (16 citations)
- AutoRedTeamer: Autonomous Red Teaming with Lifelong Attack Integration (11 citations)
- Absolute Zero: Reinforced Self-play Reasoning with Zero Data (7 citations)
- Mulberry: Empowering MLLM with o1-like Reasoning and Reflection via Collective Monte Carlo Tree Search (6 citations)
- BioReason: Incentivizing Multimodal Biological Reasoning within a DNA-LLM Model (5 citations)
- MathArena: Evaluating LLMs on Uncontaminated Math Competitions (5 citations)
- Multi-SWE-bench: A Multilingual Benchmark for Issue Resolving (5 citations)
- ContextAgent: Context-Aware Proactive LLM Agents with Open-world Sensory Perceptions (4 citations)
- SuperGPQA: Scaling LLM Evaluation across 285 Graduate Disciplines (4 citations)
- WebThinker: Empowering Large Reasoning Models with Deep Research Capability (4 citations)
- A2Seek: Towards Reasoning-Centric Benchmark for Aerial Anomaly Understanding (3 citations)
- AgentRecBench: Benchmarking LLM Agent-based Personalized Recommender Systems (3 citations)
and 186 more in the interactive map.
Instruction and fine-tuning · 109 papers
- Large Language Diffusion Models (17 citations)
- VideoRFT: Incentivizing Video Reasoning Capability in MLLMs via Reinforced Fine-Tuning (6 citations)
- ToolRL: Reward is All Tool Learning Needs (4 citations)
- Hyperbolic Fine-Tuning for Large Language Models (3 citations)
- Self-Adapting Language Models (3 citations)
- Bilevel ZOFO: Efficient LLM Fine-Tuning and Meta-Training (2 citations)
- Brain-Informed Fine-Tuning for Improved Multilingual Understanding in Language Models (2 citations)
- ChemOrch: Empowering LLMs with Chemical Intelligence via Groundbreaking Synthetic Instructions (2 citations)
- DeepVideo-R1: Video Reinforcement Fine-Tuning via Difficulty-aware Regressive GRPO (2 citations)
- GoRA: Gradient-driven Adaptive Low Rank Adaptation (2 citations)
- Improving Data Efficiency for LLM Reinforcement Fine-tuning Through Difficulty-targeted Online Data Selection and Rollout Replay (2 citations)
- Mechanism Design for LLM Fine-tuning with Multiple Reward Models (2 citations)
and 97 more in the interactive map.
AI and society · 61 papers
- Position: Towards Bidirectional Human-AI Alignment (16 citations)
- TabArena: A Living Benchmark for Machine Learning on Tabular Data (11 citations)
- LLM Generated Persona is a Promise with a Catch (10 citations)
- Utility Engineering: Analyzing and Controlling Emergent Value Systems in AIs (9 citations)
- The Leaderboard Illusion (4 citations)
- AI-Researcher: Autonomous Scientific Innovation (3 citations)
- Position: Benchmarking is Broken - Don't Let AI be Its Own Judge (3 citations)
- Rigor in AI: Doing Rigorous AI Work Requires a Broader, Responsible AI-Informed Conception of Rigor (3 citations)
- Machine Unlearning Doesn't Do What You Think: Lessons for Generative AI Policy and Research (2 citations)
- Real-Time Hyper-Personalized Generative AI Should Be Regulated to Prevent the Rise of "Digital Heroin" (2 citations)
- ResearchCodeBench: Benchmarking LLMs on Implementing Novel Machine Learning Research Code (2 citations)
- Stop DDoS Attacking the Research Community with AI-Generated Survey Papers (2 citations)
and 49 more in the interactive map.
Alignment and preferences · 59 papers
- VTON-VLLM: Aligning Virtual Try-On Models with Human Preferences (5 citations)
- Less is More: Improving LLM Alignment via Preference Data Selection (4 citations)
- Mix Data or Merge Models? Balancing the Helpfulness, Honesty, and Harmlessness of Large Language Model via Model Merging (3 citations)
- Simplicity Prevails: Rethinking Negative Preference Optimization for LLM Unlearning (3 citations)
- HelpSteer3-Preference: Open Human-Annotated Preference Data across Diverse Tasks and Languages (2 citations)
- Inference-Time Reward Hacking in Large Language Models (2 citations)
- KL-Regularized RLHF with Multiple Reference Models: Exact Solutions and Sample Complexity (2 citations)
- Aligning Compound AI Systems via System-level DPO (1 citation)
- BLEUBERI: BLEU is a surprisingly effective reward for instruction following (1 citation)
- Checklists Are Better Than Reward Models For Aligning Language Models (1 citation)
- Implicit Reward as the Bridge: A Unified View of SFT and DPO Connections (1 citation)
- OptiScene: LLM-driven Indoor Scene Layout Generation via Scaled Human-aligned Data Synthesis and Multi-Stage Preference Optimization (1 citation)
and 47 more in the interactive map.
Retrieval and knowledge · 55 papers
- Chain-of-Retrieval Augmented Generation (5 citations)
- Diagnosing and Addressing Pitfalls in KG-RAG Datasets: Toward More Reliable Benchmarking (2 citations)
- Iterative Self-Incentivization Empowers Large Language Models as Agentic Searchers (2 citations)
- Worse than Zero-shot? A Fact-Checking Dataset for Evaluating the Robustness of RAG Against Misleading Retrievals (2 citations)
- A$^3$E: Towards Compositional Model Editing (1 citation)
- Conflict-Aware Knowledge Editing in the Wild: Semantic-Augmented Graph Representation for Unstructured Text (1 citation)
- Conformal Information Pursuit for Interactively Guiding Large Language Models (1 citation)
- Deliberation on Priors: Trustworthy Reasoning of Large Language Models on Knowledge Graphs (1 citation)
- DynamicRAG: Leveraging Outputs of Large Language Model as Feedback for Dynamic Reranking in Retrieval-Augmented Generation (1 citation)
- HawkBench: Investigating Resilience of RAG Methods on Stratified Information-Seeking Tasks (1 citation)
- Influence Guided Context Selection for Effective Retrieval-Augmented Generation (1 citation)
- Process vs. Outcome Reward: Which is Better for Agentic RAG Reinforcement Learning (1 citation)
and 43 more in the interactive map.