23
Behavior Foundations for Quadruped Robots: ABot-C0 Technical Report
The ABot-C0 technical report details a new framework for quadrupedal robot control designed to overcome the scarcity of animal motion-capture data.
Top stories in Research
23
Persuasion Attacks Can Decrease Effectiveness of CoT Monitoring
Researchers have found that persuasion-based jailbreak attacks can undermine the effectiveness of chain-of-thought monitoring in detecting misaligned AI behavior.
23
TRACE: A Two-Channel Robust Attribution Watermark via Complementary Embeddings for LLM-Agent Trajectories
Researchers developed TRACE, a robust watermarking method designed to protect and verify the attribution of LLM agent trajectory logs against unauthorized rebranding or model substitution.
23
VocaDet: Sample-Driven Open-Vocabulary Object Detection and Segmentation via Visual Tokenization and Vector Database Retrieval
VocaDet is a new open-vocabulary object detection and segmentation framework that scales to large object categories using visual tokenization and vector database retrieval.
23
Linear Attention Architectures: Mechanisms, Trade-offs, and Cross-Layer Routing
This paper compares softmax attention with four recurrent linear-attention architectures to analyze their mechanisms and trade-offs for long-context processing.
23
ParamMute: Suppressing Knowledge-Critical FFNs for Faithful Retrieval-Augmented Generation
Researchers have introduced ParamMute, a method that improves the faithfulness of retrieval-augmented generation by suppressing knowledge-critical feed-forward networks that cause models to ignore retrieved context.
23
GitLake: Git-for-data for the agentic lakehouse
GitLake introduces a version-control system for data lakehouses that allows autonomous AI agents to work on isolated branches before merging changes.
23
Echoes: A semantically-aligned music deepfake detection dataset
The Echoes dataset provides over 130 hours of audio across multiple genres to benchmark and train robust music deepfake detection systems.
23
TAG: A Lightweight Framework for Test-Driven Agentic Artifact Generation
The paper introduces TAG, a lightweight framework designed to improve the reliability of LLM-generated structured artifacts through rigorous, test-driven validation.
23
Bridging Modal Isolation in Interleaved Thinking: Supervising Modality Transitions via Stepwise Reinforcement
This paper proposes a stepwise reinforcement learning approach to prevent text and image modalities from diverging during complex, interleaved multimodal reasoning tasks.
23
LoKA: Low-precision Kernel Applications for Recommendation Models At Scale
The LoKA framework introduces low-precision FP8 kernels tailored to the numerical sensitivities and communication demands of large-scale recommendation models.
23
Multimodal Unlearning Across Vision, Language, Video, and Audio: Survey of Methods, Datasets, and Benchmarks
This paper provides a comprehensive survey of methods, datasets, and benchmarks for multimodal machine unlearning across various data modalities.
23
Open-ended Multi-agent Autocurricula via Visual Inspection of Policies with Multi-modal LLMs
This study introduces a method that uses multimodal language models to visually evaluate reinforcement learning agent behaviors to design more effective training curricula.
23
Adaptive Generation of Bias-Eliciting Questions for LLMs
Researchers developed a method to adaptively generate bias-eliciting questions to better identify and evaluate inherent biases in large language models.
23
Different Teachers, Different Capabilities: Sub-1B On-Device Distillation for Structured Text Enrichment
Researchers evaluated the effectiveness of distilling structured text extraction capabilities from an 8-billion parameter reasoning model into a sub-1-billion parameter on-device model.
23
What LLM Forecasters Know but Don't Say: Probing Internal Representations for Calibration and Faithfulness
Probing the internal activations of forecasting language models reveals that their hidden representations contain more accurate and better-calibrated probability estimates than their generated text.
23
The Illusion of Equivalency: Statistical Characterization of Quantization Effects in LLMs
A new study proposes a decision-level metric called correctness agreement to better capture the behavioral changes in large language models caused by post-training quantization.
23
A First-Principles Theory of Slow Thinking and Active Perception
This paper provides a mathematical formulation of slow thinking and active perception to guide the design and training of reasoning-focused language models.
23
Dual-Difficulty Curriculum Learning for Direct Preference Optimization
Researchers introduced a two-dimensional curriculum learning framework that categorizes alignment difficulty by prompt complexity and pairwise distinguishability to optimize direct preference optimization.
23
Deployment-Time Memorization in Foundation-Model Agents
This research analyzes how deployment-time memory configurations in foundation-model agents affect personalization utility, data extraction risks, and deletion fidelity.
23
WCog-VLA: A Dual-Level World-Cognitive Vision-Language-Action Model for End-to-End Autonomous Driving
The WCog-VLA framework integrates semantic world forecasting and generative world evolution to enable proactive decision-making in end-to-end autonomous driving.
23
Predicting Scale-Up of Metal-Organic Framework Syntheses with Large Language Models
Researchers developed ScaleMOF, a dataset and fine-tuning strategy that enables large language models to predict and prioritize viable candidates for scaling up metal-organic framework syntheses.
23
A Stochastic--Geometric Theory of Scaling Laws in Grokking
This study provides a theoretical characterization of the mechanisms behind delayed generalization, or grokking, in neural networks using a stochastic-geometric framework.
23
Agentic Neural Architecture Search
Researchers propose a framework that combines large language model generation with neural architecture search to automate the design of neural networks in open-ended spaces.