Sources

Language Model Alignment in Multilingual Trolley Problems5 findings
A Paradigm Shift in Machine Translation: Boosting Translation Performance of Large Language Models4 findings
Adaptive Chameleon or Stubborn Sloth: Revealing the Behavior of Large Language Models in Knowledge Conflicts4 findings
Are Models Biased on Text without Gender-related Language?4 findings
Benchmarking Agentic Workflow Generation4 findings
Beyond Memorization: Violating Privacy via Inference with Large Language Models4 findings
Beyond task performance: evaluating and reducing the flaws of large multimodal models with in-context-learning4 findings
Boosting the visual interpretability of CLIP via adversarial fine-tuning4 findings
CAS: A Probability-Based Approach for Universal Condition Alignment Score4 findings
CS-Bench: A Comprehensive Benchmark for Large Language Models towards Computer Science Mastery4 findings
Can Knowledge Editing Really Correct Hallucinations?4 findings
Can LLMs Keep a Secret? Testing Privacy Implications of Language Models via Contextual Integrity Theory4 findings
Circuit Representation Learning with Masked Gate Modeling and Verilog-AIG Alignment4 findings
CofCA: A STEP-WISE Counterfactual Multi-hop QA benchmark4 findings
Correlating instruction-tuning (in multimodal models) with vision-language processing (in the brain)4 findings
DOCS: Quantifying Weight Similarity for Deeper Insights into Large Language Models4 findings
DailyDilemmas: Revealing Value Preferences of LLMs with Quandaries of Daily Life4 findings
Discovering Influential Neuron Path in Vision Transformers4 findings
Do LLMs Recognize Your Preferences? Evaluating Personalized Preference Following in LLMs4 findings
Does Progress On Object Recognition Benchmarks Improve Generalization on Crowdsourced, Global Data?4 findings
Episodic Memories Generation and Evaluation Benchmark for Large Language Models4 findings
FairerCLIP: Debiasing CLIP's Zero-Shot Predictions using Functions in RKHSs4 findings
FaithEval: Can Your Language Model Stay Faithful to Context, Even If "The Moon is Made of Marshmallows"4 findings
Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!4 findings
First-Person Fairness in Chatbots4 findings
Geometry of Geophysical Knowledge in TerraMind's Latent Space4 findings
ICLR: In-Context Learning of Representations4 findings
Improving Instruction-Following in Language Models through Activation Steering4 findings
Inspection and Control of Self-Generated-Text Recognition Ability in Llama3-8b-Instruct4 findings
Interpreting CLIP's Image Representation via Text-Based Decomposition4 findings
KITAB: Evaluating LLMs on Constraint Satisfaction for Information Retrieval4 findings
LLMs Know More Than They Show: On the Intrinsic Representation of LLM Hallucinations4 findings
LOKI: A Comprehensive Synthetic Data Detection Benchmark using Large Multimodal Models4 findings
Learning Performance-Improving Code Edits4 findings
Linear Representations of Political Perspective Emerge in Large Language Models4 findings
MMDT: Decoding the Trustworthiness and Safety of Multimodal Foundation Models4 findings
MMWorld: Towards Multi-discipline Multi-faceted World Model Evaluation in Videos4 findings
MediConfusion: Can you trust your AI radiologist? Probing the reliability of multimodal medical foundation models4 findings
MetaTool Benchmark for Large Language Models: Deciding Whether to Use Tools and Which to Use4 findings
MiniGPT-4: Enhancing Vision-Language Understanding with Advanced Large Language Models4 findings
Monitoring Latent World States in Language Models with Propositional Probes4 findings
More RLHF, More Trust? On The Impact of Preference Alignment On Trustworthiness4 findings
Not All Language Model Features Are One-Dimensionally Linear4 findings
Number Cookbook: Number Understanding of Language Models and How to Improve It4 findings
On Calibration of LLM-based Guard Models for Reliable Content Moderation4 findings
On the Analysis of GAN-based Image-to-Image Translation with Gaussian Noise Injection4 findings
On the Role of Attention Heads in Large Language Model Safety4 findings
One Hundred Neural Networks and Brains Watching Videos: Lessons from Alignment4 findings
Procedural Knowledge in Pretraining Drives Reasoning in Large Language Models4 findings
ReCogLab: a framework testing relational reasoning & cognitive hypotheses on LLMs4 findings
Revisiting In-context Learning Inference Circuit in Large Language Models4 findings
Robotouille: An Asynchronous Planning Benchmark for LLM Agents4 findings
SKILL-MIX: a Flexible and Expandable Family of Evaluations for AI Models4 findings
SORRY-Bench: Systematically Evaluating Large Language Model Safety Refusal4 findings
SOTOPIA: Interactive Evaluation for Social Intelligence in Language Agents4 findings
ShortcutsBench: A Large-Scale Real-world Benchmark for API-based Agents4 findings
Structure Language Models for Protein Conformation Generation4 findings
Successor Heads: Recurring, Interpretable Attention Heads In The Wild4 findings
SysBench: Can LLMs Follow System Message?4 findings
TRACE: Temporal Grounding Video LLM via Causal Event Modeling4 findings
Talk like a Graph: Encoding Graphs for Large Language Models4 findings
The "Law'' of the Unconscious Contrastive Learner: Probabilistic Alignment of Unpaired Modalities4 findings
The Generative AI Paradox: “What It Can Create, It May Not Understand”4 findings
The Hidden Language of Diffusion Models4 findings
To Trust or Not to Trust? Enhancing Large Language Models' Situated Faithfulness to External Contexts4 findings
Towards Understanding Factual Knowledge of Large Language Models4 findings
Training-Free Activation Sparsity in Large Language Models4 findings
Transformer Block Coupling and its Correlation with Generalization in LLMs4 findings
Two Effects, One Trigger: On the Modality Gap, Object Bias, and Information Imbalance in Contrastive Vision-Language Models4 findings
Understanding In-Context Learning from Repetitions4 findings
Understanding and Mitigating Hallucination in Large Vision-Language Models via Modular Attribution and Intervention4 findings
Unearthing Skill-level Insights for Understanding Trade-offs of Foundation Models4 findings
VOILA: Evaluation of MLLMs For Perceptual Understanding and Analogical Reasoning4 findings
Vision-by-Language for Training-Free Compositional Image Retrieval4 findings
What does the Knowledge Neuron Thesis Have to do with Knowledge?4 findings
A Benchmark for Learning to Translate a New Language from One Grammar Book3 findings
AI Sandbagging: Language Models can Strategically Underperform on Evaluations3 findings
An Image is Worth More Than 16x16 Patches: Exploring Transformers on Individual Pixels3 findings
Answer, Assemble, Ace: Understanding How LMs Answer Multiple Choice Questions3 findings
Arithmetic Without Algorithms: Language Models Solve Math with a Bag of Heuristics3 findings
Attention Satisfies: A Constraint-Satisfaction Lens on Factual Errors of Language Models3 findings
Auditing Sybil: Explaining Deep Lung Cancer Risk Prediction Through Generative Interventional Attributions3 findings
BadEdit: Backdooring Large Language Models by Model Editing3 findings
BigDocs: An Open Dataset for Training Multimodal Models on Document and Code Tasks3 findings
Bongard-OpenWorld: Few-Shot Reasoning for Free-form Visual Concepts in the Real World3 findings
Bootstrapping Variational Information Pursuit with Large Language and Vision Models for Interpretable Image Classification3 findings
COCO-Periph: Bridging the Gap Between Human and Machine Perception in the Periphery3 findings
COLLIE: Systematic Construction of Constrained Text Generation Tasks3 findings
Can LLM-Generated Misinformation Be Detected?3 findings
Can LLMs Express Their Uncertainty? An Empirical Evaluation of Confidence Elicitation in LLMs3 findings
ChatQA 2: Bridging the Gap to Proprietary LLMs in Long Context and RAG Capabilities3 findings
Circuit Component Reuse Across Tasks in Transformer Language Models3 findings
Compressing LLMs: The Truth is Rarely Pure and Never Simple3 findings
Confidence Elicitation: A New Attack Vector for Large Language Models3 findings
Conversational Drug Editing Using Retrieval and Domain Feedback3 findings
DENEVIL: TOWARDS DECIPHERING AND NAVIGATING THE ETHICAL VALUES OF LARGE LANGUAGE MODELS VIA INSTRUCTION LEARNING3 findings
Differentially Private Steering for Large Language Model Alignment3 findings
Discovering Failure Modes of Text-guided Diffusion Models via Adversarial Search3 findings
Distributional Associations vs In-Context Reasoning: A Study of Feed-forward and Attention Layers3 findings
Do I Know This Entity? Knowledge Awareness and Hallucinations in Language Models3 findings
Do LLMs have Consistent Values?3 findings
Do Vision & Language Decoders use Images and Text equally? How Self-consistent are their Explanations?3 findings
Do as We Do, Not as You Think: the Conformity of Large Language Models3 findings
Dual Process Learning: Controlling Use of In-Context vs. In-Weights Strategies with Weight Forgetting3 findings
DynaMath: A Dynamic Visual Benchmark for Evaluating Mathematical Reasoning Robustness of Vision Language Models3 findings
EFFICIENT JAILBREAK ATTACK SEQUENCES ON LARGE LANGUAGE MODELS VIA MULTI-ARMED BANDIT-BASED CONTEXT SWITCHING3 findings
EffoVPR: Effective Foundation Model Utilization for Visual Place Recognition3 findings
Endless Jailbreaks with Bijection Learning3 findings
FIXLIP: Explaining Similarity in Vision-Language Encoders with Weighted Banzhaf Interactions3 findings
FairMT-Bench: Benchmarking Fairness for Multi-turn Dialogue in Conversational LLMs3 findings
Fine-Tuning Enhances Existing Mechanisms: A Case Study on Entity Tracking3 findings
Follow My Instruction and Spill the Beans: Scalable Data Extraction from Retrieval-Augmented Generation Systems3 findings
Frozen Transformers in Language Models Are Effective Visual Encoder Layers3 findings
GPT-4 Is Too Smart To Be Safe: Stealthy Chat with LLMs via Cipher3 findings
Glimpse: Enabling White-Box Methods to Use Proprietary Models for Zero-Shot LLM-Generated Text Detection3 findings
Harnessing Explanations: LLM-to-LM Interpreter for Enhanced Text-Attributed Graph Representation Learning3 findings
Have the VLMs Lost Confidence? A Study of Sycophancy in VLMs3 findings
How do Language Models Bind Entities in Context?3 findings
INViTE: INterpret and Control Vision-Language Models with Text Explanations3 findings
In-Context Learning Dynamics with Random Binary Sequences3 findings
In-Context Learning Learns Label Relationships but Is Not Conventional Learning3 findings
InsightBench: Evaluating Business Analytics Agents Through Multi-Step Insight Generation3 findings
Interpretable Diffusion via Information Decomposition3 findings
Interpreting and Editing Vision-Language Representations to Mitigate Hallucinations3 findings
Interpreting the Second-Order Effects of Neurons in CLIP3 findings
Is Self-Repair a Silver Bullet for Code Generation?3 findings
Is Your Video Language Model a Reliable Judge?3 findings
JoMA: Demystifying Multilayer Transformers via Joint Dynamics of MLP and Attention3 findings
Knowledge Localization: Mission Not Accomplished? Enter Query Localization!3 findings
LMSYS-Chat-1M: A Large-Scale Real-World LLM Conversation Dataset3 findings
Language Modeling Is Compression3 findings
Language Models Are Implicitly Continuous3 findings
Large Language Model Cascades with Mixture of Thought Representations for Cost-Efficient Reasoning3 findings
Large Language Models Cannot Self-Correct Reasoning Yet3 findings
Lawma: The Power of Specialization for Legal Annotation3 findings
Linearity of Relation Decoding in Transformer Language Models3 findings
LitCab: Lightweight Language Model Calibration over Short- and Long-form Responses3 findings
Look Before You Leap: Universal Emergent Mechanism for Retrieval in Language Models3 findings
MINT: Evaluating LLMs in Multi-turn Interaction with Tools and Language Feedback3 findings
Measuring Vision-Language STEM Skills of Neural Models3 findings
Neuron-based Multifractal Analysis of Neuron Interaction Dynamics in Large Models3 findings
NoVo: Norm Voting off Hallucinations with Attention Heads in Large Language Models3 findings
OBI-Bench: Can LMMs Aid in Study of Ancient Script on Oracle Bones?3 findings
OctoPack: Instruction Tuning Code Large Language Models3 findings
On the Humanity of Conversational AI: Evaluating the Psychological Portrayal of LLMs3 findings
On the self-verification limitations of large language models on reasoning and planning tasks3 findings
OpenRCA: Can Large Language Models Locate the Root Cause of Software Failures?3 findings
PandaLM: An Automatic Evaluation Benchmark for LLM Instruction Tuning Optimization3 findings
PerceptionCLIP: Visual Classification by Inferring and Conditioning on Contexts3 findings
Perturbation-Restrained Sequential Model Editing3 findings
Phenomenal Yet Puzzling: Testing Inductive Reasoning Capabilities of Language Models with Hypothesis Refinement3 findings
Physics of Language Models: Part 3.2, Knowledge Manipulation3 findings
Prometheus: Inducing Fine-Grained Evaluation Capability in Language Models3 findings
Protein Language Model Fitness is a Matter of Preference3 findings
Quantifying Generalization Complexity for Large Language Models3 findings
Refining CLIP's Spatial Awareness: A Visual-Centric Perspective3 findings
Rephrase, Augment, Reason: Visual Grounding of Questions for Vision-Language Models3 findings
Robust LLM safeguarding via refusal feature adversarial training3 findings
SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesis3 findings
SEAL: A Framework for Systematic Evaluation of Real-World Super-Resolution3 findings
Safety Alignment Should be Made More Than Just a Few Tokens Deep3 findings
ScImage: How good are multimodal large language models at scientific text-to-image generation?3 findings
Scaling and evaluating sparse autoencoders3 findings
Solving Challenging Math Word Problems Using GPT-4 Code Interpreter with Code-based Self-Verification3 findings
Sparse Autoencoders Find Highly Interpretable Features in Language Models3 findings
Sparse Autoencoders Reveal Temporal Difference Learning in Large Language Models3 findings
Sparse Feature Circuits: Discovering and Editing Interpretable Causal Graphs in Language Models3 findings
Supervised Knowledge Makes Large Language Models Better In-context Learners3 findings
Tell Your Model Where to Attend: Post-hoc Attention Steering for LLMs3 findings
The 3D-PC: a benchmark for visual perspective taking in humans and machines3 findings
The Alignment Problem from a Deep Learning Perspective3 findings
The Cost of Scaling Down Large Language Models: Reducing Model Size Affects Memory before In-context Learning3 findings
The Devil is in the Object Boundary: Towards Annotation-free Instance Segmentation using Foundation Models3 findings
The Same but Different: Structural Similarities and Differences in Multilingual Language Modeling3 findings
Towards Interpreting Visual Information Processing in Vision-Language Models3 findings
Training on the Test Task Confounds Evaluation and Emergence3 findings
Uncovering Latent Memories in Large Language Models3 findings
Understanding Catastrophic Forgetting in Language Models via Implicit Inference3 findings
Understanding In-Context Learning in Transformers and LLMs by Learning to Learn Discrete Functions3 findings
Understanding and Enhancing the Transferability of Jailbreaking Attacks3 findings
UniversalNER: Targeted Distillation from Large Language Models for Open Named Entity Recognition3 findings
Vector-ICL: In-context Learning with Continuous Vector Representations3 findings
VibeCheck: Discover and Quantify Qualitative Differences in Large Language Models3 findings
Vision-Language Models are Zero-Shot Reward Models for Reinforcement Learning3 findings
What Makes Large Language Models Reason in (Multi-Turn) Code Generation?3 findings
What does the TerraMind model look at and think about?3 findings
Where and Why TerraMind's Optical-to-SAR Generation Fails3 findings
WildBench: Benchmarking LLMs with Challenging Tasks from Real Users in the Wild3 findings
WorkflowLLM: Enhancing Workflow Orchestration Capability of Large Language Models3 findings
A Geometric Framework for Understanding Memorization in Generative Models2 findings
AI as Humanity’s Salieri: Quantifying Linguistic Creativity of Language Models via Systematic Attribution of Machine Text against Web Text2 findings
Analyzing Feed-Forward Blocks in Transformers through the Lens of Attention Maps2 findings
AutoBencher: Towards Declarative Benchmark Construction2 findings
Beyond Single Concept Vector: Modeling Concept Subspace in LLMs with Gaussian Distribution2 findings
Beyond Surface Structure: A Causal Assessment of LLMs' Comprehension ability2 findings
Bias Runs Deep: Implicit Reasoning Biases in Persona-Assigned LLMs2 findings
Bisimulation Metric for Model Predictive Control2 findings
BooookScore: A systematic exploration of book-length summarization in the era of LLMs2 findings
CRITIC: Large Language Models Can Self-Correct with Tool-Interactive Critiquing2 findings
Calibrating Expressions of Certainty2 findings
Can In-context Learning Really Generalize to Out-of-distribution Tasks?2 findings
Can Sensitive Information Be Deleted From LLMs? Objectives for Defending Against Extraction Attacks2 findings
Can We Talk Models Into Seeing the World Differently?2 findings
Capability Localization: Capabilities Can be Localized rather than Individual Knowledge2 findings
Catastrophic Jailbreak of Open-source LLMs via Exploiting Generation2 findings
Causal Order: The Key to Leveraging Imperfect Experts in Causal Inference2 findings
Closing the Curious Case of Neural Text Degeneration2 findings
Competition Dynamics Shape Algorithmic Phases of In-Context Learning2 findings
Concept vectors in TerraMind latent space2 findings
ConceptPrune: Concept Editing in Diffusion Models via Skilled Neuron Pruning2 findings
Convolution Meets LoRA: Parameter Efficient Finetuning for Segment Anything Model2 findings
Cross the Gap: Exposing the Intra-modal Misalignment in CLIP via Modality Inversion2 findings
DICE: End-to-end Deformation Capture of Hand-Face Interactions from a Single Image2 findings
Davidsonian Scene Graph: Improving Reliability in Fine-grained Evaluation for Text-to-Image Generation2 findings
Debiasing Algorithm through Model Adaptation2 findings
Dissecting learning and forgetting in language model finetuning2 findings
Do LLMs ``know'' internally when they follow instructions?2 findings
Does Refusal Training in LLMs Generalize to the Past Tense?2 findings
Does Safety Training of LLMs Generalize to Semantically Related Natural Prompts?2 findings
Efficient Streaming Language Models with Attention Sinks2 findings
Eliminating Position Bias of Language Models: A Mechanistic Approach2 findings
Evaluating Large Language Models through Role-Guide and Self-Reflection: A Comparative Study2 findings
Facing the Elephant in the Room: Visual Prompt Tuning or Full finetuning?2 findings
Faithful and Efficient Explanations for Neural Networks via Neural Tangent Kernel Surrogate Models2 findings
Function Vectors in Large Language Models2 findings
GAIA: a benchmark for General AI Assistants2 findings
GNNX-BENCH: Unravelling the Utility of Perturbation-based GNN Explainers through In-depth Benchmarking2 findings
How Does Vision-Language Adaptation Impact the Safety of Vision Language Models?2 findings
Improving Reasoning Performance in Large Language Models via Representation Engineering2 findings
Injecting Universal Jailbreak Backdoors into LLMs in Minutes2 findings
Input Space Mode Connectivity in Deep Neural Networks2 findings
InterpGNN: Understand and Improve Generalization Ability of Transdutive GNNs through the Lens of Interplay between Train and Test Nodes2 findings
Interpretable Vision-Language Survival Analysis with Ordinal Inductive Bias for Computational Pathology2 findings
Is Factuality Enhancement a Free Lunch For LLMs? Better Factuality Can Lead to Worse Context-Faithfulness2 findings
Is This the Subspace You Are Looking for? An Interpretability Illusion for Subspace Activation Patching2 findings
Jailbreak in pieces: Compositional Adversarial Attacks on Multi-Modal Language Models2 findings
Jailbreaking Leading Safety-Aligned LLMs with Simple Adaptive Attacks2 findings
Judge Decoding: Faster Speculative Sampling Requires Going Beyond Model Alignment2 findings
Knowledge Entropy Decay during Language Model Pretraining Hinders New Knowledge Acquisition2 findings
LLM-wrapper: Black-Box Semantic-Aware Adaptation of Vision-Language Models for Referring Expression Comprehension2 findings
LLaMA-Omni: Seamless Speech Interaction with Large Language Models2 findings
Language Model Cascades: Token-Level Uncertainty And Beyond2 findings
Language Models Represent Space and Time2 findings
Large Language Models Assume People are More Rational than We Really are2 findings
Large Language Models as Automated Aligners for benchmarking Vision-Language Models2 findings
Large Language Models can Become Strong Self-Detoxifiers2 findings
Large-scale and Fine-grained Vision-language Pre-training for Enhanced CT Image Understanding2 findings
Layer-wise linear mode connectivity2 findings
Learning Dynamics of LLM Finetuning2 findings
Lines of Thought in Large Language Models2 findings
Localizing and Editing Knowledge In Text-to-Image Generative Models2 findings
MIRAGE: Evaluating and Explaining Inductive Reasoning Process in Language Models2 findings
Matcher: Segment Anything with One Shot Using All-Purpose Feature Matching2 findings
Measuring and Enhancing Trustworthiness of LLMs in RAG through Grounded Attributions and Learning to Refuse2 findings
Memorization Capacity of Multi-Head Attention in Transformers2 findings
Model Editing as a Robust and Denoised variant of DPO: A Case Study on Toxicity2 findings
Monet: Mixture of Monosemantic Experts for Transformers2 findings
MuSR: Testing the Limits of Chain-of-thought with Multistep Soft Reasoning2 findings
NL-Eye: Abductive NLI For Images2 findings
Network Memory Footprint Compression Through Jointly Learnable Codebooks and Mappings2 findings
Neuron based Personality Trait Induction in Large Language Models2 findings
On Linear Representations and Pretraining Data Frequency in Language Models2 findings
Parsing neural dynamics with infinite recurrent switching linear dynamical systems2 findings
Quantifying Language Models' Sensitivity to Spurious Features in Prompt Design or: How I learned to start worrying about prompt formatting2 findings
ReLU Strikes Back: Exploiting Activation Sparsity in Large Language Models2 findings
Residual Stream Analysis with Multi-Layer SAEs2 findings
Rethinking the Benefits of Steerable Features in 3D Equivariant Graph Neural Networks2 findings
Robustness of AI-Image Detectors: Fundamental Limits and Practical Attacks2 findings
RocketEval: Efficient automated LLM evaluation via grading checklist2 findings
SaMer: A Scenario-aware Multi-dimensional Evaluator for Large Language Models2 findings
Safety-Tuned LLaMAs: Lessons From Improving the Safety of Large Language Models that Follow Instructions2 findings
Sparse autoencoders reveal selective remapping of visual concepts during adaptation2 findings
Style Outweighs Substance: Failure Modes of LLM Judges in Alignment Benchmarking2 findings
Sudden Drops in the Loss: Syntax Acquisition, Phase Transitions, and Simplicity Bias in MLMs2 findings
Sufficient Context: A New Lens on Retrieval Augmented Generation Systems2 findings
Synergy and Diversity in CLIP: Enhancing Performance Through Adaptive Backbone Ensembling2 findings
Teaching Large Language Models to Self-Debug2 findings
The Semantic Hub Hypothesis: Language Models Share Semantic Representations Across Languages and Modalities2 findings
The Trickle-down Impact of Reward Inconsistency on RLHF2 findings
The Unlocking Spell on Base LLMs: Rethinking Alignment via In-Context Learning2 findings
ToRA: A Tool-Integrated Reasoning Agent for Mathematical Problem Solving2 findings
Towards Neural Scaling Laws for Time Series Foundation Models2 findings
Towards Principled Evaluations of Sparse Autoencoders for Interpretability and Control2 findings
Towards Universality: Studying Mechanistic Similarity Across Language Model Architectures2 findings
Tracing Representation Progression: Analyzing and Enhancing Layer-Wise Similarity2 findings
Turning large language models into cognitive models2 findings
Understanding and Enhancing Safety Mechanisms of LLMs via Safety-Specific Neuron2 findings
Understanding prompt engineering may not require rethinking generalization2 findings
Unveiling the Pitfalls of Knowledge Editing for Large Language Models2 findings
VLM2Vec: Training Vision-Language Models for Massive Multimodal Embedding Tasks2 findings
Vanishing Gradients in Reinforcement Finetuning of Language Models2 findings
ViLMA: A Zero-Shot Benchmark for Linguistic and Temporal Grounding in Video-Language Models2 findings
Vision Transformers Need Registers2 findings
Walk the Talk? Measuring the Faithfulness of Large Language Model Explanations2 findings
WebRL: Training LLM Web Agents via Self-Evolving Online Curriculum Reinforcement Learning2 findings
What Secrets Do Your Manifolds Hold? Understanding the Local Geometry of Generative Models2 findings
When Do Prompting and Prefix-Tuning Work? A Theory of Capabilities and Limitations2 findings
Where Are We Now? Investigating Spatial Information in TerraMind's Latent Space2 findings
Your Weak LLM is Secretly a Strong Teacher for Alignment2 findings
Zero and Few-shot Semantic Parsing with Ambiguous Inputs2 findings
h4rm3l: A Language for Composable Jailbreak Attack Synthesis2 findings
$\mathcal{B}$-Coder: Value-Based Deep Reinforcement Learning for Program Synthesis1 finding
A Formal Framework for Understanding Length Generalization in Transformers1 finding
A Simple Interpretable Transformer for Fine-Grained Image Classification and Analysis1 finding
A Theoretical Analysis of Self-Supervised Learning for Vision Transformers1 finding
A Theory for Token-Level Harmonization in Retrieval-Augmented Generation1 finding
A path-norm toolkit for modern networks: consequences, promises and challenges1 finding
ADIFF: Explaining audio difference using natural language1 finding
AHA: A Vision-Language-Model for Detecting and Reasoning Over Failures in Robotic Manipulation1 finding
API Pack: A Massive Multi-Programming Language Dataset for API Call Generation1 finding
AlpaGasus: Training a Better Alpaca with Fewer Data1 finding
Alt-Text with Context: Improving Accessibility for Images on Twitter1 finding
Are Bert Family Good Instruction Followers? A Study on Their Potential And Limitations1 finding
Asynchronous RLHF: Faster and More Efficient Off-Policy RL for Language Models1 finding
At Which Training Stage Does Code Data Help LLMs Reasoning?1 finding
BadChain: Backdoor Chain-of-Thought Prompting for Large Language Models1 finding
Bayesian Regularization of Latent Representation1 finding
Beyond correlation: The impact of human uncertainty in measuring the effectiveness of automatic evaluation and LLM-as-a-judge1 finding
Beyond single neurons: population response geometry in digital twins of mouse visual cortex1 finding
Black-Box Detection of Language Model Watermarks1 finding
BlueSuffix: Reinforced Blue Teaming for Vision-Language Models Against Jailbreak Attacks1 finding
Brain decoding: toward real-time reconstruction of visual perception1 finding
Breaking Free from MMI: A New Frontier in Rationalization by Probing Input Utilization1 finding
Circumventing Concept Erasure Methods For Text-To-Image Generative Models1 finding
Concept Bottleneck Generative Models1 finding
Concept Bottleneck Language Models For Protein Design1 finding
Context Clues: Evaluating Long Context Models for Clinical Prediction Tasks on EHR Data1 finding
Context-Parametric Inversion: Why Instruction Finetuning May Not Actually Improve Context Reliance1 finding
Contrastive Learning is Spectral Clustering on Similarity Graph1 finding
Controllable Context Sensitivity and the Knob Behind It1 finding
Cross-Entropy Is All You Need To Invert the Data Generating Process1 finding
Data-independent Module-aware Pruning for Hierarchical Vision Transformers1 finding
Decoding Natural Images from EEG for Object Recognition1 finding
Defining and extracting generalizable interaction primitives from DNNs1 finding
Demystifying Embedding Spaces using Large Language Models1 finding
Density estimation with LLMs: a geometric investigation of in-context learning trajectories1 finding
Detecting Backdoor Samples in Contrastive Language Image Pretraining1 finding
Differentiation and Specialization of Attention Heads via the Refined Local Learning Coefficient1 finding
Does CLIP’s generalization performance mainly stem from high train-test similarity?1 finding
Does Writing with Language Models Reduce Content Diversity?1 finding
Dynamic-LLaVA: Efficient Multimodal Large Language Models via Dynamic Vision-language Context Sparsification1 finding
Efficient Dynamics Modeling in Interactive Environments with Koopman Theory1 finding
Elucidating the Exposure Bias in Diffusion Models1 finding
Elucidating the design space of classifier-guided diffusion generation1 finding
Emergence of a High-Dimensional Abstraction Phase in Language Transformers1 finding
Enhancing Transferable Adversarial Attacks on Vision Transformers through Gradient Normalization Scaling and High-Frequency Adaptation1 finding
Entropy is not Enough for Test-Time Adaptation: From the Perspective of Disentangled Factors1 finding
Explaining Modern Gated-Linear RNNs via a Unified Implicit Attention Formulation1 finding
Exploiting Distribution Constraints for Scalable and Efficient Image Retrieval1 finding
Failures to Find Transferable Image Jailbreaks Between Vision-Language Models1 finding
Faithful Rule Extraction for Differentiable Rule Learning Models1 finding
FakeShield: Explainable Image Forgery Detection and Localization via Multi-modal Large Language Models1 finding
Feature Averaging: An Implicit Bias of Gradient Descent Leading to Non-Robustness in Neural Networks1 finding
FreSh: Frequency Shifting for Accelerated Neural Representation Learning1 finding
From Bricks to Bridges: Product of Invariances to Enhance Latent Space Communication1 finding
GReaTer: Gradients Over Reasoning Makes Smaller Language Models Strong Prompt Optimizers1 finding
Generative Verifiers: Reward Modeling as Next-Token Prediction1 finding
Gramian Multimodal Representation Learning and Alignment1 finding
HaDeMiF: Hallucination Detection and Mitigation in Large Language Models1 finding
Harnessing Webpage UIs for Text-Rich Visual Understanding1 finding
How new data permeates LLM knowledge and how to dilute it1 finding
How to Fine-Tune Vision Models with SGD1 finding
Human-Aligned Chess With a Bit of Search1 finding
HyperDAS: Towards Automating Mechanistic Interpretability with Hypernetworks1 finding
ImpScore: A Learnable Metric For Quantifying The Implicitness Level of Sentences1 finding
Improving Complex Reasoning with Dynamic Prompt Corruption: A Soft Prompt Optimization Approach1 finding
Improving Long-Text Alignment for Text-to-Image Diffusion Models1 finding
Improving protein optimization with smoothed fitness landscapes1 finding
In Search of Forgotten Domain Generalization1 finding
Is Large-scale Pretraining the Secret to Good Domain Generalization?1 finding
Is attention required for ICL? Exploring the Relationship Between Model Architecture and In-Context Learning Ability1 finding
LLM-based Typed Hyperresolution for Commonsense Reasoning with Knowledge Bases1 finding
LLM-grounded Video Diffusion Models1 finding
Label-Focused Inductive Bias over Latent Object Features in Visual Classification1 finding
Language-Informed Visual Concept Learning1 finding
LanguageBind: Extending Video-Language Pretraining to N-modality by Language-based Semantic Alignment1 finding
LeanAgent: Lifelong Learning for Formal Theorem Proving1 finding
Learning LLM-as-a-Judge for Preference Alignment1 finding
Learning to Contextualize Web Pages for Enhanced Decision Making by LLM Agents1 finding
Locality Alignment Improves Vision-Language Models1 finding
Logicbreaks: A Framework for Understanding Subversion of Rule-based Inference1 finding
MIntRec2.0: A Large-scale Benchmark Dataset for Multimodal Intent Recognition and Out-of-scope Detection in Conversations1 finding
MMKE-Bench: A Multimodal Editing Benchmark for Diverse Visual Knowledge1 finding
Massive Editing for Large Language Models via Meta Learning1 finding
Mechanism and Emergence of Stacked Attention Heads in Multi-Layer Transformers1 finding
Mechanistic Permutability: Match Features Across Layers1 finding
Multi-granularity Correspondence Learning from Long-term Noisy Videos1 finding
NetInfoF Framework: Measuring and Exploiting Network Usable Information1 finding
NeurFlow: Interpreting Neural Networks through Neuron Groups and Functional Interactions1 finding
Neuron Activation Coverage: Rethinking Out-of-distribution Detection and Generalization1 finding
On Evaluating the Durability of Safeguards for Open-Weight LLMs1 finding
On the Foundations of Shortcut Learning1 finding
On the Hölder Stability of Multiset and Graph Neural Networks1 finding
On the Variance of Neural Network Training with respect to Test Sets and Distributions1 finding
Open-YOLO 3D: Towards Fast and Accurate Open-Vocabulary 3D Instance Segmentation1 finding
Overthinking the Truth: Understanding how Language Models Process False Demonstrations1 finding
PRIME: Prioritizing Interpretability in Failure Mode Extraction1 finding
Peering Through Preferences: Unraveling Feedback Acquisition for Aligning Large Language Models1 finding
PlaSma: Procedural Knowledge Models for Language-based Planning and Re-Planning1 finding
Poison-splat: Computation Cost Attack on 3D Gaussian Splatting1 finding
Post-hoc Reward Calibration: A Case Study on Length Bias1 finding
Precise Parameter Localization for Textual Generation in Diffusion Models1 finding
Predicting Emergent Abilities with Infinite Resolution Evaluation1 finding
Predictive, scalable and interpretable knowledge tracing on structured domains1 finding
Probabilistic Adaptation of Black-Box Text-to-Video Models1 finding
Probe before You Talk: Towards Black-box Defense against Backdoor Unalignment for Large Language Models1 finding
Provably Safeguarding a Classifier from OOD and Adversarial Samples1 finding
QPM: Discrete Optimization for Globally Interpretable Image Classification1 finding
Quantifying the Plausibility of Context Reliance in Neural Machine Translation1 finding
RAPPER: Reinforced Rationale-Prompted Paradigm for Natural Language Explanation in Visual Question Answering1 finding
RT-Trajectory: Robotic Task Generalization via Hindsight Trajectory Sketches1 finding
Ranking-aware adapter for text-driven image ordering with CLIP1 finding
ReDeEP: Detecting Hallucination in Retrieval-Augmented Generation via Mechanistic Interpretability1 finding
Reassessing How to Compare and Improve the Calibration of Machine Learning Models1 finding
Reclaiming the Source of Programmatic Policies: Programmatic versus Latent Spaces1 finding
Reconstructive Visual Instruction Tuning1 finding
Respect the model: Fine-grained and Robust Explanation with Sharing Ratio Decomposition1 finding
Rethinking Self-Distillation: Label Averaging and Enhanced Soft Label Refinement with Partial Labels1 finding
Retrieval Head Mechanistically Explains Long-Context Factuality1 finding
Retrieval meets Long Context Large Language Models1 finding
Ring-A-Bell! How Reliable are Concept Removal Methods For Diffusion Models?1 finding
SWE-bench: Can Language Models Resolve Real-world Github Issues?1 finding
SaRA: High-Efficient Diffusion Model Fine-tuning with Progressive Sparse Low-Rank Adaptation1 finding
Scaling for Training Time and Post-hoc Out-of-distribution Detection Enhancement1 finding
Simple is Effective: The Roles of Graphs and Large Language Models in Knowledge-Graph-Based Retrieval-Augmented Generation1 finding
Skip-Attention: Improving Vision Transformers by Paying Less Attention1 finding
Stable Segment Anything Model1 finding
Tailoring Self-Rationalizers with Multi-Reward Distillation1 finding
Teaching Arithmetic to Small Transformers1 finding
Temporal Reasoning Transfer from Text to Video1 finding
The Devil is in the Neurons: Interpreting and Mitigating Social Biases in Language Models1 finding
The False Promise of Imitating Proprietary Language Models1 finding
The Geometry of Categorical and Hierarchical Concepts in Large Language Models1 finding
The Power of LLM-Generated Synthetic Data for Stance Detection in Online Political Discussions1 finding
The Reasonableness Behind Unreasonable Translation Capability of Large Language Model1 finding
The Reversal Curse: LLMs trained on “A is B” fail to learn “B is A”1 finding
TiC-CLIP: Continual Training of CLIP Models1 finding
To the Cutoff... and Beyond? A Longitudinal Perspective on LLM Data Contamination1 finding
ToolLLM: Facilitating Large Language Models to Master 16000+ Real-world APIs1 finding
Toward Understanding In-context vs. In-weight Learning1 finding
Towards 3D Molecule-Text Interpretation in Language Models1 finding
Towards Best Practices of Activation Patching in Language Models: Metrics and Methods1 finding
Towards Certification of Uncertainty Calibration under Adversarial Attacks1 finding
Towards Robust Multi-Modal Reasoning via Model Selection1 finding
Towards Understanding Sycophancy in Language Models1 finding
Towards Understanding Text Hallucination of Diffusion Models via Local Generation Bias1 finding
Transformers Learn Low Sensitivity Functions: Investigations and Implications1 finding
TypedThinker: Diversify Large Language Model Reasoning with Typed Thinking1 finding
U-shaped and Inverted-U Scaling behind Emergent Abilities of Large Language Models1 finding
Uncovering Gaps in How Humans and LLMs Interpret Subjective Language1 finding
Understanding Augmentation-based Self-Supervised Representation Learning via RKHS Approximation and Regression1 finding
Understanding and Mitigating the Label Noise in Pre-training on Downstream Tasks1 finding
Understanding the Robustness of Randomized Feature Defense Against Query-Based Adversarial Attacks1 finding
Unveiling and Manipulating Prompt Influence in Large Language Models1 finding
Unveiling the Unseen: Identifiable Clusters in Trained Depthwise Convolutional Kernels1 finding
What Makes a Good Prune? Maximal Unstructured Pruning for Maximal Cosine Similarity1 finding
Where We Have Arrived in Proving the Emergence of Sparse Interaction Primitives in DNNs1 finding
WildChat: 1M ChatGPT Interaction Logs in the Wild1 finding
YOLO-RD: Introducing Relevant and Compact Explicit Knowledge to YOLO by Retriever-Dictionary1 finding
YaRN: Efficient Context Window Extension of Large Language Models1 finding