DC
Danqi Chen
cs.CLcs.LGcs.AIcs.CRcs.IRcs.CYcs.CVstat.MLAIalignment
On Valency
published · living versionsW_r955csby·v1 · currentpublished
Suppressing Misaligned Behaviors in Language Models, Durably
with Luxi He, Pengcheng Jiang, Jifan Zhang, Jiawei Han
1 version
W_ftvgr5v6·v1published
Preprints & journals
99 papers in the corpus · 2013–2026Where RL Breaks: Localizing and Repairing Failures via Targeted Continued Pretraining2610.25836v1 · Xinyi Wang, Danqi Chen2026 · 0 citationsarXivon Valency
Language Models that Play Chess and Explain Their Moves2610.03695v1 · Adithya Bhaskar, Jeffrey Cheng, Danqi Chen2026 · 0 citationsarXiv
Statutory construction and interpretation for AI.42475576 · He, Luxi, Nadeem, Nimra, Liao, Michel et al.2026 · 0 citationsProceedings of the National Academy of Sciences of the United States of America. 2026;123(30):e2509766123
Hubs or Fringes: Pretraining Data Selection via Web Graph Centrality2606.11499v1 · Vedant Badoni, Danqi Chen, Xinyi Wang2026 · 0 citationsarXiv
Goedel-Architect: Streamlining Formal Theorem Proving with Blueprint Generation and Refinement2606.06468v1 · Jui-Hui Chung, Ziyang Cai, Zihao Li et al.2026 · 0 citationsarXiv
Odysseus: Scaling VLMs to 100+ Turn Decision-Making in Games via Reinforcement Learning2605.00347v1 · Chengshuai Shi, Wenzhe Li, Xinran Liang et al.2026 · 0 citationsarXiv
Agentic Aggregation for Parallel Scaling of Long-Horizon Agentic Tasks2604.11753v3 · Yoonsang Lee, Howard Yen, Xi Ye et al.2026 · 0 citationsarXiv
Self-Distillation Zero: Self-Revision Turns Binary Rewards into Dense Supervision2604.12002v2 · Yinghui He, Simran Kaur, Adithya Bhaskar et al.2026 · 0 citationsarXiv
Vero: An Open RL Recipe for General Visual Reasoning2604.04917v3 · Gabriel Sarch, Linrong Cai, Qunzhong Wang et al.2026 · 0 citationsarXiv
DySCO: Dynamic Attention-Scaling Decoding for Long-Context Language Models2602.22175v2 · Xi Ye, Wuwei Zhang, Fangcong Yin et al.2026 · 0 citationsarXiv
Retaining by Doing: The Role of On-Policy Data in Mitigating Forgetting2510.18874v3 · Howard Chen, Noam Razin, Karthik Narasimhan et al.2025 · 0 citationsProceedings of the 43rd International Conference on Machine Learning (ICML), 2026
Lost in the Maze: Overcoming Context Limitations in Long-Horizon Agentic Search2510.18939v2 · Howard Yen, Yoonsang Lee, Ashwin Paranjape et al.2025 · 0 citationsarXiv
Extracting Rule-based Descriptions of Attention Features in Transformers2510.18148v1 · Dan Friedman, Adithya Bhaskar, Alexander Wettig et al.2025 · 0 citationsarXiv
Language Models that Think, Chat Better2509.20357v2 · Adithya Bhaskar, Xi Ye, Danqi Chen2025 · 0 citationsarXiv
Statutory Construction and Interpretation for Artificial Intelligence2509.01186v1 · Luxi He, Nimra Nadeem, Michel Liao et al.2025 · 1 citationarXiv
Goedel-Prover-V2: Scaling Formal Theorem Proving with Scaffolded Data Synthesis and Self-Correction2508.03613v1 · Yong Lin, Shange Tang, Bohan Lyu et al.2025 · 0 citationsarXiv
Cache Me If You Can: How Many KVs Do You Need for Effective Long-Context LMs?2506.17121v1 · Adithya Bhaskar, Alexander Wettig, Tianyu Gao et al.2025 · 0 citationsarXiv
Query-Focused Retrieval Heads Improve Long-Context Reasoning and Re-ranking2506.09944v2 · Wuwei Zhang, Fangcong Yin, Howard Yen et al.2025 · 1 citationarXiv
Precise Information Control in Long-Form Text Generation2506.06589v2 · Jacqueline He, Howard Yen, Margaret Li et al.2025 · 0 citationsarXiv
The Surprising Effectiveness of Negative Reinforcement in LLM Reasoning2506.01347v2 · Xinyu Zhu, Mengzhou Xia, Zhepei Wei et al.2025 · 3 citationsarXiv
The Model Hears You: Audio Language Model Deployments Should Consider the Principle of Least Privilege2503.16833v2 · Luxi He, Xiangyu Qi, Michel Liao et al.2025 · 0 citationsarXiv
Organize the Web: Constructing Domains Enhances Pre-Training Data Curation2502.10341v3 · Alexander Wettig, Kyle Lo, Sewon Min et al.2025 · 0 citationsarXiv
Goedel-Prover: A Frontier Model for Open-Source Automated Theorem Proving2502.07640v3 · Yong Lin, Shange Tang, Bohan Lyu et al.2025 · 3 citationsarXiv
LongProc: Benchmarking Long-Context Language Models on Long Procedural Generation2501.05414v3 · Xi Ye, Fangcong Yin, Yinghui He et al.2025 · 1 citationarXiv
Metadata Conditioning Accelerates Language Model Pre-training2501.01956v3 · Tianyu Gao, Alexander Wettig, Luxi He et al.2025 · 2 citationsarXiv
Continual Memorization of Factoids in Language Models2411.07175v3 · Howard Chen, Jiayi Geng, Adithya Bhaskar et al.2024 · 0 citationsTransactions on Machine Learning Research, 2026
Unintentional Unalignment: Likelihood Displacement in Direct Preference Optimization2410.08847v4 · Noam Razin, Sadhika Malladi, Adithya Bhaskar et al.2024 · 0 citationsarXiv
How to Train Long-Context Language Models (Effectively)2410.02660v4 · Tianyu Gao, Alexander Wettig, Howard Yen et al.2024 · 6 citationsarXiv
HELMET: How to Evaluate Long-Context Language Models Effectively and Thoroughly2410.02694v3 · Howard Yen, Tianyu Gao, Minmin Hou et al.2024 · 2 citationsarXiv
BRIGHT: A Realistic and Challenging Benchmark for Reasoning-Intensive Retrieval2407.12883v4 · Hongjin Su, Howard Yen, Mengzhou Xia et al.2024 · 5 citationsarXiv
Representing Rule-based Chatbots with Transformers2407.10949v2 · Dan Friedman, Abhishek Panigrahi, Danqi Chen2024 · 0 citationsarXiv
LitSearch: A Retrieval Benchmark for Scientific Literature Search2407.18940v2 · Anirudh Ajith, Mengzhou Xia, Alexis Chevalier et al.2024 · 13 citationsarXiv
CharXiv: Charting Gaps in Realistic Chart Understanding in Multimodal LLMs2406.18521v1 · Zirui Wang, Mengzhou Xia, Luxi He et al.2024 · 28 citationsarXiv
Finding Transformer Circuits with Edge Pruning2406.16778v3 · Adithya Bhaskar, Alexander Wettig, Dan Friedman et al.2024 · 8 citationsarXiv
Fantastic Copyrighted Beasts and How (Not) to Generate Them2406.14526v2 · Luxi He, Yangsibo Huang, Weijia Shi et al.2024 · 3 citationsarXiv
SORRY-Bench: Systematically Evaluating Large Language Model Safety Refusal2406.14598v2 · Tinghao Xie, Xiangyu Qi, Yi Zeng et al.2024 · 7 citationsarXiv
AI Risk Management Should Incorporate Both Safety and Security2405.19524v1 · Xiangyu Qi, Yangsibo Huang, Yi Zeng et al.2024 · 3 citationsarXiv
Certifiably Robust RAG against Retrieval Corruption2405.15556v2 · Chong Xiang, Tong Wu, Zexuan Zhong et al.2024 · 7 citationsarXiv
SimPO: Simple Preference Optimization with a Reference-Free Reward2405.14734v3 · Yu Meng, Mengzhou Xia, Danqi Chen2024 · 61 citationsarXiv
Lory: Fully Differentiable Mixture-of-Experts for Autoregressive Language Model Pre-training2405.03133v2 · Zexuan Zhong, Mengzhou Xia, Danqi Chen et al.2024 · 7 citationsarXiv
Foundational Challenges in Assuring Alignment and Safety of Large Language Models2404.09932v2 · Usman Anwar, Abulhair Saparov, Javier Rando et al.2024 · 19 citationsarXiv
The Heuristic Core: Understanding Subnetwork Generalization in Pretrained Language Models2403.03942v2 · Adithya Bhaskar, Dan Friedman, Danqi Chen2024 · 1 citationarXiv
Reliable, Adaptable, and Attributable Language Models with Retrieval2403.03187v1 · Akari Asai, Zexuan Zhong, Danqi Chen et al.2024 · 9 citationsarXiv
Long-Context Language Modeling with Parallel Context Encoding2402.16617v2 · Howard Yen, Tianyu Gao, Danqi Chen2024 · 9 citationsarXiv
Improving Language Understanding from Screenshots2402.14073v1 · Tianyu Gao, Zirui Wang, Adithya Bhaskar et al.2024 · 0 citationsarXiv
Language Models as Science Tutors2402.11111v2 · Alexis Chevalier, Jiayi Geng, Alexander Wettig et al.2024 · 8 citationsarXiv
QuRating: Selecting High-Quality Data for Training Language Models2402.09739v3 · Alexander Wettig, Aatmik Gupta, Saumya Malik et al.2024 · 6 citationsarXiv
LESS: Selecting Influential Data for Targeted Instruction Tuning2402.04333v3 · Mengzhou Xia, Sadhika Malladi, Suchin Gururangan et al.2024 · 19 citationsarXiv
Interpretability Illusions in the Generalization of Simplified Models2312.03656v2 · Dan Friedman, Andrew Lampinen, Lucas Dixon et al.2023 · 1 citationarXiv
Poisoning Retrieval Corpora by Injecting Adversarial Passages2310.19156v1 · Zexuan Zhong, Ziqing Huang, Alexander Wettig et al.2023 · 47 citationsarXiv
Career total: 162 works. 99 are in this corpus.Showing the 50 most recent.
Profile built from the corpus for this byline.
Author records are still filling in while the Hub is in alpha. If this is your page, you'll be able to claim it soon. Spot a mistake? Tell us.