ICLR 2025 accepted papers on Large language models
742 of the 3,703 papers accepted at ICLR 2025 (20.0%) are on Large language models, compared with 14.5% at ICLR 2024. They fall into 7 topics.
See ICLR 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 · 240 papers
- Physics of Language Models: Part 3.3, Knowledge Capacity Scaling Laws (13 citations)
- NV-Embed: Improved Techniques for Training LLMs as Generalist Embedding Models (10 citations)
- Justice or Prejudice? Quantifying Biases in LLM-as-a-Judge (9 citations)
- Improving Instruction-Following in Language Models through Activation Steering (8 citations)
- Spider 2.0: Evaluating Language Models on Real-World Enterprise Text-to-SQL Workflows (7 citations)
- NatureLM-audio: an Audio-Language Foundation Model for Bioacoustics (6 citations)
- SORRY-Bench: Systematically Evaluating Large Language Model Safety Refusal (6 citations)
- Beyond Model Collapse: Scaling Up with Synthesized Data Requires Verification (5 citations)
- Prompting Fairness: Integrating Causality to Debias Large Language Models (5 citations)
- Jailbreak Antidote: Runtime Safety-Utility Balance via Sparse Representation Adjustment in Large Language Models (4 citations)
- LLMs Know More Than They Show: On the Intrinsic Representation of LLM Hallucinations (4 citations)
- Perplexed by Perplexity: Perplexity-Based Data Pruning With Small Reference Models (4 citations)
and 228 more in the interactive map.
Agents, code and math · 134 papers
- GSM-Symbolic: Understanding the Limitations of Mathematical Reasoning in Large Language Models (53 citations)
- Can LLMs Generate Novel Research Ideas? A Large-Scale Human Study with 100+ NLP Researchers (44 citations)
- LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Code (31 citations)
- WizardMath: Empowering Mathematical Reasoning for Large Language Models via Reinforced Evol-Instruct (28 citations)
- Mixture-of-Agents Enhances Large Language Model Capabilities (26 citations)
- BigCodeBench: Benchmarking Code Generation with Diverse Function Calls and Complex Instructions (22 citations)
- IRIS: LLM-Assisted Static Analysis for Detecting Security Vulnerabilities (17 citations)
- AFlow: Automating Agentic Workflow Generation (7 citations)
- CHASE-SQL: Multi-Path Reasoning and Preference Optimized Candidate Selection in Text-to-SQL (7 citations)
- Efficient Evolutionary Search Over Chemical Space with Large Language Models (6 citations)
- LLM-SR: Scientific Equation Discovery via Programming with Large Language Models (6 citations)
- ScienceAgentBench: Toward Rigorous Assessment of Language Agents for Data-Driven Scientific Discovery (6 citations)
and 122 more in the interactive map.
Alignment and preferences · 97 papers
- Self-Play Preference Optimization for Language Model Alignment (11 citations)
- Eliciting Human Preferences with Language Models (9 citations)
- Magpie: Alignment Data Synthesis from Scratch by Prompting Aligned LLMs with Nothing (6 citations)
- HelpSteer2-Preference: Complementing Ratings with Preferences (4 citations)
- Language Models Learn to Mislead Humans via RLHF (4 citations)
- On Targeted Manipulation and Deception when Optimizing LLMs for User Feedback (4 citations)
- WildBench: Benchmarking LLMs with Challenging Tasks from Real Users in the Wild (4 citations)
- Do LLMs Recognize Your Preferences? Evaluating Personalized Preference Following in LLMs (3 citations)
- JudgeBench: A Benchmark for Evaluating LLM-Based Judges (3 citations)
- Taming Overconfidence in LLMs: Reward Calibration in RLHF (3 citations)
- Towards Federated RLHF with Aggregated Client Preference for LLMs (3 citations)
- Trust or Escalate: LLM Judges with Provable Guarantees for Human Agreement (3 citations)
and 85 more in the interactive map.
Retrieval and knowledge · 91 papers
- Generative Representational Instruction Tuning (15 citations)
- Think-on-Graph 2.0: Deep and Faithful Large Language Model Reasoning with Knowledge-guided Retrieval Augmented Generation (9 citations)
- SiReRAG: Indexing Similar and Related Information for Multihop Reasoning (8 citations)
- Looking Inward: Language Models Can Learn About Themselves by Introspection (7 citations)
- Physics of Language Models: Part 3.2, Knowledge Manipulation (7 citations)
- Speculative RAG: Enhancing Retrieval Augmented Generation through Drafting (7 citations)
- Long-Context LLMs Meet RAG: Overcoming Challenges for Long Inputs in RAG (6 citations)
- VisRAG: Vision-based Retrieval-augmented Generation on Multi-modality Documents (6 citations)
- ReDeEP: Detecting Hallucination in Retrieval-Augmented Generation via Mechanistic Interpretability (5 citations)
- Simple is Effective: The Roles of Graphs and Large Language Models in Knowledge-Graph-Based Retrieval-Augmented Generation (5 citations)
- Accelerating Inference of Retrieval-Augmented Generation via Sparse Context Selection (4 citations)
- ALLaM: Large Language Models for Arabic and English (4 citations)
and 79 more in the interactive map.
Instruction and fine-tuning · 88 papers
- JudgeLM: Fine-tuned Large Language Models are Scalable Judges (18 citations)
- Does Refusal Training in LLMs Generalize to the Past Tense? (7 citations)
- NeuroLM: A Universal Multi-task Foundation Model for Bridging the Gap between Language and EEG Signals (7 citations)
- Generative Monoculture in Large Language Models (6 citations)
- Is In-Context Learning Sufficient for Instruction Following in LLMs? (4 citations)
- Preserving Diversity in Supervised Fine-Tuning of Large Language Models (4 citations)
- Safety Alignment Should be Made More Than Just a Few Tokens Deep (4 citations)
- Self-play with Execution Feedback: Improving Instruction-following Capabilities of Large Language Models (4 citations)
- Encryption-Friendly LLM Architecture (3 citations)
- Multimodal Large Language Models for Inverse Molecular Design with Retrosynthetic Planning (3 citations)
- REEF: Representation Encoding Fingerprints for Large Language Models (3 citations)
- ROUTE: Robust Multitask Tuning and Collaboration for Text-to-SQL (3 citations)
and 76 more in the interactive map.
Reasoning and chain of thought · 67 papers
- To CoT or not to CoT? Chain-of-thought helps mainly on math and symbolic reasoning (13 citations)
- Advancing LLM Reasoning Generalists with Preference Trees (5 citations)
- Mind the Gap: Examining the Self-Improvement Capabilities of Large Language Models (5 citations)
- On the self-verification limitations of large language models on reasoning and planning tasks (3 citations)
- Physics of Language Models: Part 2.2, How to Learn From Mistakes on Grade-School Math Problems (3 citations)
- Semantics-Adaptive Activation Intervention for LLMs via Dynamic Steering Vectors (3 citations)
- Dualformer: Controllable Fast and Slow Thinking by Learning with Randomized Reasoning Traces (2 citations)
- Generative Verifiers: Reward Modeling as Next-Token Prediction (2 citations)
- Multiagent Finetuning: Self Improvement with Diverse Reasoning Chains (2 citations)
- Reasoning with Latent Thoughts: On the Power of Looped Transformers (2 citations)
- B-STaR: Monitoring and Balancing Exploration and Exploitation in Self-Taught Reasoners (1 citation)
- Better Instruction-Following Through Minimum Bayes Risk (1 citation)
and 55 more in the interactive map.
AI and society · 25 papers
- AI Sandbagging: Language Models can Strategically Underperform on Evaluations (5 citations)
- NNsight and NDIF: Democratizing Access to Open-Weight Foundation Model Internals (4 citations)
- CycleResearcher: Improving Automated Research via Automated Review (2 citations)
- DailyDilemmas: Revealing Value Preferences of LLMs with Quandaries of Daily Life (2 citations)
- Language Models Trained to do Arithmetic Predict Human Risky and Intertemporal Choice (2 citations)
- Large Language Models Assume People are More Rational than We Really are (2 citations)
- Measuring And Improving Persuasiveness Of Large Language Models (2 citations)
- AnalogGenie: A Generative Engine for Automatic Discovery of Analog Circuit Topologies (1 citation)
- Benchmarking Predictive Coding Networks -- Made Simple (1 citation)
- ForecastBench: A Dynamic Benchmark of AI Forecasting Capabilities (1 citation)
- Human Simulacra: Benchmarking the Personification of Large Language Models (1 citation)
- Human-Aligned Chess With a Bit of Search (1 citation)
and 13 more in the interactive map.