DK
Daniel Khashabi
cs.CLcs.AIcs.LGcs.CVcs.ROcs.CYstat.MLcs.HCcs.IRcs.SE
On Valency
published · living versionsW_s6bdne4e·v1 · currentpublished
Training Foundation Models with a Single Algorithm
with Yuda Song, Alvin Zhang, Fahim Tajwar, Daman Arora +5
1 version
Preprints & journals
128 papers in the corpus · 2012–2026Accurate but Not Humble: Evaluating Epistemic Humility in LLM Agents under Knowledge Conflict2610.12360v1 · Kaiser Sun, Bernal Jimenez Gutierrez, Hongjun Liu et al.2026 · 0 citationsEMNLP2026
CrossWeave: Bridging Perspectives Across Online Communities with a Dual-Pane Design2610.10441v1 · Fei Fang, Reva Hirave, William Jurayj et al.2026 · 0 citationsCompanion of the Computer-Supported Cooperative Work and Social Computing (CSCW Companion 2026)
SIGMA: Self-Improving Alignment Generalization from a Model Spec2610.07935v1 · Jingyu Zhang, Shruti Palaskar, Daniel Khashabi et al.2026 · 0 citationsarXiv
JEPA-TTT: Persistent Test-Time Training of Latent World Models for Planning under Dynamics Shifts2610.00722v1 · Zheyuan Zhang, Suyu Ye, Nakul Agarwal et al.2026 · 0 citationsarXiv
Harness Learning Enables Generalizable Test-Time Adaptation2609.35738v1 · Alvin Zhang, Xuecheng Liu, Zixuan Wang et al.2026 · 0 citationsarXiv
Minimally Invasive Steering of Language Models2609.30218v1 · Taha Entesari, Jingyu Zhang, Daniel Khashabi et al.2026 · 0 citationsarXiv
Convergent Emergence of In-Context Learning Across Modalities2609.14011v1 · Nathan Breslow, Seungwook Han, Daniel Hyunsoo Lee et al.2026 · 0 citationsarXiv
Continual Learning Mechanisms Compose for Long-Horizon Memorization2609.06986v1 · Zheyuan Zhang, Alvin Zhang, Daniel Khashabi et al.2026 · 0 citationsarXiv
VBVR-Pro: A Scalable and Verifiable Suite for Native Visual Reasoning2608.26105v2 · Junxiang Xu, Ruisi Wang, Fanyi Pu et al.2026 · 0 citationsarXiv
Information Abundance Paradox: Long-Context Training Undermines Parametric Knowledge2608.12218v2 · Arda Uzunoglu, Benjamin Van Durme, Daniel Khashabi2026 · 0 citationsarXiv
Better, Faster, Stronger: Programmatic Skill Learning Best Reduces Agent Cost2608.11338v1 · Zixi Huang, Xiheng Wang, Andrew Wang et al.2026 · 0 citationsarXiv
Self-Compacting Language Model Agents2606.23525v2 · Tianjian Li, Jingyu Zhang, William Jurayj et al.2026 · 0 citationsarXiv
Trust Functions: Near-Lossless Weak-to-Strong Generalization by Learning When to Trust the Weak Teacher2606.01000v1 · Arda Uzunoglu, Alvin Zhang, Daniel Khashabi2026 · 0 citationsarXiv
AgentOdyssey: Open-Ended Long-Horizon Text Game Generation for Test-Time Continual Learning Agents2606.24893v2 · Zheyuan Zhang, Zehao Wen, Alvin Zhang et al.2026 · 0 citationsarXiv
How to Interpret Agent Behavior2605.13625v1 · Jie Gao, Kaiser Sun, Jen-tse Huang et al.2026 · 0 citationsarXiv
Can Coding Agents Reproduce Findings in Computational Materials Science?2605.00803v2 · Ziyang Huang, Yi Cao, Ali K. Shargh et al.2026 · 0 citationsarXiv
Many-Tier Instruction Hierarchy in LLM Agents2604.09443v4 · Jingyu Zhang, Tianjian Li, William Jurayj et al.2026 · 0 citationsarXiv
Steered LLM Activations are Non-Surjective2604.09839v2 · Aayush Mishra, Daniel Khashabi, Anqi Liu2026 · 0 citationsarXiv
GENFIG1: Visual Summaries of Scholarly Work as a Challenge for Vision-Language Models2604.04172v1 · Yaohan Guan, Pristina Wang, Najim Dehak et al.2026 · 0 citationsarXiv
Are Finer Citations Always Better? Rethinking Granularity for Attributed Generation2604.01432v4 · Hexuan Wang, Jingyu Zhang, Benjamin Van Durme et al.2026 · 0 citationsarXiv
Crystal: Characterizing Relative Impact of Scholarly Publications2603.26791v3 · Hannah Collison, Benjamin Van Durme, Daniel Khashabi2026 · 0 citationsarXiv
A Very Big Video Reasoning Suite2602.20159v4 · Maijunxian Wang, Ruisi Wang, Juyi Lin et al.2026 · 0 citationsarXiv
Safe and Interpretable Multimodal Path Planning for Multi-Agent Cooperation2602.19304v1 · Haojun Shi, Suyu Ye, Katherine M. Guerrerio et al.2026 · 0 citationsarXiv
Conformal Thinking: Risk Control for Reasoning on a Compute Budget2602.03814v2 · Xi Wang, Anushri Suresh, Alvin Zhang et al.2026 · 0 citationsarXiv
More Than Efficiency: Embedding Compression Improves Domain Adaptation in Dense Retrieval2601.13525v3 · Chunsheng Zuo, Daniel Khashabi2026 · 2 citationsarXiv
Principled Context Engineering for RAG: Statistical Guarantees via Conformal Prediction2511.17908v2 · Debashish Chakraborty, Eugene Yang, Daniel Khashabi et al.2025 · 1 citationarXiv
Genomic Next-Token Predictors are In-Context Learners2511.12797v3 · Nathan Breslow, Aayush Mishra, Mahler Revsine et al.2025 · 0 citationsarXiv
CreativityPrism: A Cross-Domain Evaluation Framework for Large Language Model Creativity2510.20091v3 · Zhaoyi Joey Hou, Bowei Alvin Zhang, Yining Lu et al.2025 · 0 citationsarXiv
Query Decomposition for RAG: Balancing Exploration-Exploitation2510.18633v1 · Roxana Petcu, Kenton Murray, Daniel Khashabi et al.2025 · 0 citationsarXiv
World-in-World: World Models in a Closed-Loop World2510.18135v2 · Jiahan Zhang, Muqing Jiang, Nanru Dai et al.2025 · 0 citationsarXiv
GOLD PANNING: Strategic Context Shuffling for Needle-in-Haystack Reasoning2510.09770v3 · Adam Byerly, Daniel Khashabi2025 · 0 citationsarXiv
The Alignment Waltz: Jointly Training Agents to Collaborate for Safety2510.08240v2 · Jingyu Zhang, Haozhu Wang, Eric Michael Smith et al.2025 · 0 citationsarXiv
BiomedSQL: Text-to-SQL for Scientific Reasoning on Biomedical Knowledge Bases.41031078 · Koretsky, Mathew J, Willey, Maya, Bianchi, Owen et al.2025 · 2 citationsArXiv. 2025
Controlling the Risk of Corrupted Contexts for Language Models via Early-Exiting2510.02480v3 · Andrea Wynn, Metod Jazbec, Charith Peris et al.2025 · 0 citationsarXiv
The Flaw of Averages: Measuring Benchmark-Level Distributional Robustness2509.25671v2 · Arda Uzunoglu, Tianjian Li, Daniel Khashabi2025 · 0 citationsarXiv
IA2: Alignment with ICL Activations Improves Supervised Fine-Tuning2509.22621v3 · Aayush Mishra, Daniel Khashabi, Anqi Liu2025 · 0 citationsarXiv
Challenging the Evaluator: LLM Sycophancy Under User Rebuttal2509.16533v1 · Sungwon Kim, Daniel Khashabi2025 · 6 citationsarXiv
Linguistic Nepotism: Trading-off Quality for Language Preference in Multilingual RAG2509.13930v3 · Dayeon Ki, Marine Carpuat, Paul McNamee et al.2025 · 0 citationsarXiv
Jointly Reinforcing Diversity and Quality in Language Model Generations2509.02534v1 · Tianjian Li, Yiming Zhang, Ping Yu et al.2025 · 2 citationsarXiv
Evaluating the Evaluators: Are readability metrics good measures of readability?2508.19221v1 · Isabel Cachola, Daniel Khashabi, Mark Dredze2025 · 4 citationsarXiv
Hell or High Water: Evaluating Agentic Recovery from External Failures2508.11027v1 · Andrew Wang, Sophia Hager, Adi Asija et al.2025 · 0 citationsarXiv
The Translation Barrier Hypothesis: Multilingual Generation with Large Language Models Suffers from Implicit Translation Failure2506.22724v2 · Niyati Bafna, Tianjian Li, Kenton Murray et al.2025 · 0 citationsarXiv
Feedback Friction: LLMs Struggle to Fully Incorporate External Feedback2506.11930v2 · Dongwei Jiang, Alvin Zhang, Andrew Wang et al.2025 · 0 citationsarXiv
Jailbreak Distillation: Renewable Safety Benchmarking2505.22037v1 · Jingyu Zhang, Ahmed Elgohary, Xiawei Wang et al.2025 · 0 citationsarXiv
Hidden in the Haystack: Smaller Needles are More Difficult for LLMs to Find2505.18148v2 · Owen Bianchi, Mathew J. Koretsky, Maya Willey et al.2025 · 0 citationsarXiv
Lost in the Haystack: Smaller Needles are More Difficult for LLMs to Find.41031075 · Bianchi, Owen, Koretsky, Mathew J, Willey, Maya et al.2025 · 0 citationsArXiv. 2025
SIMPLEMIX: Frustratingly Simple Mixing of Off- and On-policy Data in Language Model Preference Learning2505.02363v1 · Tianjian Li, Daniel Khashabi2025 · 0 citationsarXiv
ICL CIPHERS: Quantifying "Learning" in In-Context Learning via Substitution Ciphers2504.19395v2 · Zhouxiang Fang, Aayush Mishra, Muhan Gao et al.2025 · 0 citationsarXiv
Certified Mitigation of Worst-Case LLM Copyright Infringement2504.16046v2 · Jingyu Zhang, Jiacan Yu, Marc Marone et al.2025 · 0 citationsarXiv
From Papers to Panoramas: Building Hierarchies of Scientific Literature at Scale2504.13834v7 · Muhan Gao, Jash Shah, Weiqi Wang et al.2025 · 0 citationsarXiv
Career total: 194 works. 128 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.