5/6/2026
Generative Interfaces for Language Models
Proposes generative interfaces paradigm to move LLM interactions beyond linear request-response format for more efficient multi-turn, information-dense, and exploratory tasks.
5/6/2026
Confident, Calibrated, or Complicit: Safety Alignment and Ideological Bias in LLM Hate Speech Detection
Study evaluates safety-aligned versus uncensored LLMs on hate speech detection, revealing trade-offs between model censoring and detection performance across political personas.
5/6/2026
Improving Factuality in LLMs via Inference-Time Knowledge Graph Construction
Method improves LLM factuality by constructing knowledge graphs at inference time instead of using unstructured text retrieval, enhancing reasoning and reducing irrelevant information influence.
5/6/2026
ALIGNS: Unlocking nomological networks in psychological measurement through a large language model
ALIGNS: LLM-based approach for building nomological networks in psychological measurement to establish construct validity.
5/6/2026
Learning Decomposed Contextual Token Representations from Pretrained and Collaborative Signals for Generative Recommendation
Method for generative recommenders learning decomposed contextual token representations combining pretrained and collaborative signals.
5/6/2026
The Coding Limits of Robust Watermarking for Generative Models
Theoretical analysis of watermarking robustness for generative models using zero-bit tamper-detection codes.
5/6/2026
Less Precise Can Be More Reliable: A Systematic Evaluation of Quantization's Impact on VLMs Beyond Accuracy
Study evaluating quantization impact on vision-language models across reliability metrics beyond accuracy for OOD detection.
5/6/2026
CoSpaDi: Compressing LLMs via Calibration-Guided Sparse Dictionary Learning
CoSpaDi: training-free LLM compression using calibration-guided sparse dictionary learning as alternative to low-rank approximations.
5/6/2026
Short window attention enables long-term memorization
Analysis of hybrid attention architectures combining local sliding window and global attention for improved long-term memory.
5/6/2026
Ergodic Risk Measures: Towards a Risk-Aware Foundation for Continual Reinforcement Learning
Ergodic risk measures framework for continual RL agents balancing knowledge retention and adaptation to new environments.
5/6/2026
Turning Drift into Constraint: Robust Reasoning Alignment in Non-Stationary Environments
Framework for multi-source reasoning alignment in MLLMs addressing concept drift in non-stationary environments.
5/6/2026
TokenChain: A Discrete Speech Chain via Semantic Token Modeling
TokenChain: fully discrete speech chain coupling semantic-token ASR with two-stage TTS for joint improvement of speech tasks.
5/6/2026
CleverCatch: A Knowledge-Guided Weak Supervision Model for Fraud Detection
CleverCatch: knowledge-guided weak supervision model for healthcare fraud detection with limited labeled data.
5/6/2026
TokenTiming: A Dynamic Alignment Method for Universal Speculative Decoding Model Pairs
TokenTiming: speculative decoding acceleration for LLM inference enabling draft and target models with different vocabularies.
5/6/2026
Reinforcement Learning from Compiler and Language Server Feedback
Method training coding agents using compiler and language server feedback as supervision signals to improve program correctness.
5/6/2026
Noise is All You Need: Solving Linear Inverse Problems by Noise Combination Sampling with Diffusion Models
Method using diffusion models for solving linear inverse problems by balancing noise integration to maintain generative quality.
5/6/2026
Implicature in Interaction: Understanding Implicature Improves Alignment in Human-LLM Interaction
Study on how understanding linguistic implicature improves LLM alignment with user intent in human-AI interaction.
5/6/2026
TetraJet-v2: Accurate NVFP4 Training for Large Language Models with Oscillation Suppression and Outlier Control
TetraJet-v2: 4-bit quantized training method for LLMs using NVFP4 format with techniques to suppress oscillation and control outliers.
5/6/2026
Re-Key-Free, Risky-Free: Adaptable Model Usage Control
Method for protecting deep neural networks from unauthorized use by enabling model usage control without embedding access keys in parameters.
5/6/2026
Pistachio: Towards Synthetic, Balanced, and Long-Form Video Anomaly Benchmarks
Synthetic benchmark dataset for video anomaly detection with improved scene diversity, balanced coverage, and temporal complexity evaluation.
5/6/2026
Intelligent Agents with Emotional Intelligence: Current Trends, Challenges, and Future Prospects
Survey of intelligent agents with emotional intelligence capabilities for human-computer interaction and affective computing systems.
5/6/2026
Aligning LLMs with Biomedical Knowledge using Balanced Fine-Tuning
Balanced fine-tuning method for aligning LLMs with biomedical knowledge by addressing uncertainty structure differences in scientific text.
5/6/2026
2D-ThermAl: Physics-Informed Framework for Thermal Analysis of Circuits using Generative AI
Physics-informed generative AI framework for thermal circuit analysis, accelerating early-stage design validation vs traditional FEM simulation.
5/6/2026
jina-vlm: Small Multilingual Vision Language Model
2.4B parameter multilingual vision-language model achieving SOTA performance on multilingual VQA with token-efficient image processing.
5/6/2026
InfiniteDiffusion: Bridging Learned Fidelity and Procedural Utility for Open-World Terrain Generation
Training-free algorithm for unbounded terrain generation using diffusion models, combining procedural noise efficiency with learned fidelity.
5/6/2026
A Neuro-Symbolic Framework for Accountability in Public-Sector AI
Neuro-symbolic explainability framework for public-sector AI systems, linking AI decisions to legal requirements using structured knowledge representation.
5/6/2026
How Prompts Move Language Model Behavior: Frames, Salience, and Construal as Semantic Control
Cognitive-semantic analysis of prompt engineering as natural-language control mechanism, using frame activation and salience to explain LLM behavior modification.
5/6/2026
Disentangling Fact from Sentiment: A Dynamic Conflict-Consensus Framework for Multimodal Fake News Detection
Framework for multimodal fake news detection using conflict-consensus approach that leverages cross-modal discrepancies instead of enforcing consistency.
5/6/2026
InstructMoLE: Instruction-Guided Mixture of Low-rank Experts for Multi-Conditional Image Generation
InstructMoLE: Parameter-efficient fine-tuning of Diffusion Transformers using Mixture of Low-rank Experts with instruction-guided routing for multi-conditional image generation.
5/6/2026
LVLM-Aided Alignment of Task-Specific Vision Models
Research using Large Vision-Language Models to align task-specific vision models with human domain knowledge, reducing spurious correlations.
5/6/2026
LinMU: Multimodal Understanding Made Linear
Using large vision-language models to improve alignment of task-specific vision models with human knowledge.
5/6/2026
OpenAI GPT-5 System Card
OpenAI GPT-5 system card describing unified architecture with fast base model, deeper reasoning model, and real-time router for task-appropriate model selection.
5/6/2026
From Laboratory to Real-World Applications: Benchmarking Agentic Code Reasoning at the Repository Level
RepoReason benchmark for evaluating agentic code reasoning at repository level with white-box diagnostics for logical consistency across interdependent file systems.
5/6/2026
False Friends in the Shell: Unveiling the Emoticon Semantic Confusion in Large Language Models
Study of emoticon semantic confusion vulnerability in LLMs where ASCII-based emoticons cause misinterpretation and unintended actions.
5/6/2026
STAGE: A Full-Screenplay Benchmark for Reasoning over Evolving Storie
STAGE benchmark for evaluating model reasoning over full-screenplay narratives, testing story comprehension, character tracking, and multi-form generation consistency.
5/6/2026
Can LLMs Compress (and Decompress)? Evaluating Code Understanding and Execution via Invertibility
RoundTripCodeEval benchmark evaluating code-LLM reasoning consistency through forward-backward execution via lossless compression algorithm round-trip fidelity.
5/6/2026
Can professional translators identify machine-generated text?
Empirical study testing whether professional translators can identify AI-generated text from ChatGPT-4o versus human authors without specialized training.
5/6/2026
LLM-VA: Resolving the Jailbreak-Overrefusal Trade-off via Vector Alignment
Method using vector alignment to resolve jailbreak-overrefusal trade-off in safety-aligned LLMs by separately encoding answer vectors and safety judgment vectors.
5/6/2026
When Iterative RAG Beats Ideal Evidence: A Diagnostic Study in Scientific Multi-hop Question Answering
Diagnostic study comparing iterative RAG with static RAG for multi-hop scientific question answering, analyzing when synchronized retrieval-reasoning loops provide benefits over single-pass approaches.
5/6/2026
Cross-Session Decoding of Neural Spiking Data via Task-Conditioned Latent Alignment
TCLA framework enables cross-session neural decoding with limited target data using task-conditioned latent alignment from source sessions.
5/6/2026
VERGE: Formal Refinement and Guidance Engine for Verifiable LLM Reasoning
VERGE neurosymbolic framework combines LLMs with SMT solvers for verification-guided reasoning, decomposing outputs into formal logic for consistency checking.
5/6/2026
CoFrGeNet: Continued Fraction Architectures for Language Generation
CoFrGeNets introduces continued fraction-inspired architecture replacing attention and feed-forward layers in Transformers for improved language generation.
5/6/2026
Gated Relational Alignment via Confidence-based Distillation for Efficient VLMs
GRACE framework unifies quantization-aware training and knowledge distillation for efficient vision-language models using information bottleneck principle.
5/6/2026
VideoGPA: Distilling Geometry Priors for 3D-Consistent Video Generation
VideoGPA adds geometric priors to video diffusion models via self-supervised preference alignment to improve 3D structural consistency in generated videos.
5/6/2026
MCP-Atlas: A Large-Scale Benchmark for Tool-Use Competency with Real MCP Servers
MCP-Atlas large-scale benchmark for evaluating LLM tool-use competency with 36 real Model Context Protocol servers, capturing real-world tool invocation complexity.
5/6/2026
STEP: Warm-Started Visuomotor Policies with Spatiotemporal Consistency Prediction
STEP warm-starts diffusion-based visuomotor policies with spatiotemporal consistency prediction to reduce inference latency for real-time robotic control.
5/6/2026
Control Reinforcement Learning: Interpretable Token-Level Steering of LLMs via Sparse Autoencoder Features
Control Reinforcement Learning trains policy to select sparse autoencoder features for interpretable token-level steering of LLMs, producing explainable intervention logs.
5/6/2026
From SFT to RL: Demystifying the Post-Training Pipeline for LLM-based Vulnerability Detection
Comprehensive investigation of post-training pipeline for LLM-based vulnerability detection, demonstrating on-policy RL with GRPO outperforms supervised fine-tuning approaches.
5/6/2026
CrispEdit: Low-Curvature Projections for Scalable Non-Destructive LLM Editing
CrispEdit scalable second-order algorithm for LLM editing that preserves general capabilities while making targeted behavior changes, treating capability preservation as explicit constraint.
5/6/2026
HistCAD: A Constraint-Aware Parametric History-Based CAD Representation, Dataset, and Benchmark with Industrial Complexity
HistCAD benchmark and dataset for parametric CAD generation with constraint preservation under edits, measuring design intent retention rather than just reconstruction fidelity.