AI Agents

NeuBird AI Raises $19.3 Million To Scale Agentic AI

[2511.06448] When AI Agents Collude Online: Financial Fraud Risks by Collaborative LLM Agents on Social Platforms

Abstract page for arXiv paper 2511.06448: When AI Agents Collude Online: Financial Fraud Risks by Collaborative LLM Agents on Social Plat...

[2510.20728] Co-Designing Quantum Codes with Transversal Diagonal Gates via Multi-Agent Systems

Abstract page for arXiv paper 2510.20728: Co-Designing Quantum Codes with Transversal Diagonal Gates via Multi-Agent Systems

[2602.16849] On the Mechanism and Dynamics of Modular Addition: Fourier Features, Lottery Ticket, and Grokking

This paper analyzes how two-layer neural networks learn to solve the modular addition task, providing insights into feature learning, tra...

[2602.16837] A Residual-Aware Theory of Position Bias in Transformers

This paper presents a residual-aware theory explaining the position bias in Transformers, revealing how residual connections prevent atte...

[2602.16823] Formal Mechanistic Interpretability: Automated Circuit Discovery with Provable Guarantees

This article presents a novel approach to automated circuit discovery in neural networks, emphasizing provable guarantees for robustness ...

[2602.16793] Escaping the Cognitive Well: Efficient Competition Math with Off-the-Shelf Models

The paper presents a novel inference pipeline that leverages off-the-shelf models to solve International Mathematical Olympiad problems e...

[2602.16787] Better Think Thrice: Learning to Reason Causally with Double Counterfactual Consistency

This paper introduces Double Counterfactual Consistency (DCC), a method for evaluating and enhancing causal reasoning in large language m...

[2602.11337] MolmoSpaces: A Large-Scale Open Ecosystem for Robot Navigation and Manipulation

MolmoSpaces introduces a large-scale open ecosystem designed for benchmarking robot navigation and manipulation, featuring over 230k dive...

[2602.07666] SoK: DARPA's AI Cyber Challenge (AIxCC): Competition Design, Architectures, and Lessons Learned

This paper analyzes DARPA's AI Cyber Challenge (AIxCC), focusing on competition design, architectural approaches of finalists, and key le...

[2602.06355] Di3PO - Diptych Diffusion DPO for Targeted Improvements in Image Generation

The paper presents Di3PO, a novel method for improving image generation in text-to-image diffusion models by efficiently creating targete...

[2602.03972] Fixed Budget is No Harder Than Fixed Confidence in Best-Arm Identification up to Logarithmic Factors

This paper explores the relationship between fixed-budget and fixed-confidence settings in best-arm identification, demonstrating that th...

[2601.08697] Auditing Student-AI Collaboration: A Case Study of Online Graduate CS Students

This study audits the collaboration between online graduate CS students and AI, exploring preferences for automation in academic tasks an...

[2601.01224] Improved Object-Centric Diffusion Learning with Registers and Contrastive Alignment

This paper presents Contrastive Object-centric Diffusion Alignment (CODA), an enhancement to object-centric learning that reduces slot en...

[2512.23482] Theory of Mind for Explainable Human-Robot Interaction

This article explores the integration of Theory of Mind (ToM) in human-robot interaction (HRI) to enhance robot interpretability and user...

[2512.22213] On the Existence and Behavior of Secondary Attention Sinks

This paper explores the concept of secondary attention sinks in machine learning models, highlighting their distinct properties and behav...

[2511.00794] Efficient Reinforcement Learning for Large Language Models with Intrinsic Exploration

This paper presents PREPO, a novel approach to enhance data efficiency in reinforcement learning for large language models by leveraging ...

[2511.00040] Semi-Supervised Preference Optimization with Limited Feedback

This paper discusses Semi-Supervised Preference Optimization (SSPO), which reduces the need for extensive labeled feedback in preference ...

[2510.15297] VERA-MH Concept Paper

The VERA-MH Concept Paper outlines an innovative framework for evaluating AI chatbots in mental health contexts, focusing on suicide risk...

[2510.14974] pi-Flow: Policy-Based Few-Step Generation via Imitation Distillation

The paper presents pi-Flow, a novel approach to few-step generation in machine learning that utilizes imitation distillation to enhance m...

[2509.11461] CareerPooler: AI-Powered Metaphorical Pool Simulation Improves Experience and Outcomes in Career Exploration

CareerPooler introduces an AI-driven metaphorical simulation for career exploration, enhancing user engagement and decision-making throug...

[2506.21039] Strict Subgoal Execution: Reliable Long-Horizon Planning in Hierarchical Reinforcement Learning

The paper presents Strict Subgoal Execution (SSE), a novel framework for hierarchical reinforcement learning that enhances long-horizon p...

[2506.20555] DeepQuark: A Deep-Neural-Network Approach to Multiquark Bound States

The paper presents DeepQuark, a novel deep-neural-network approach for analyzing multiquark bound states, demonstrating superior performa...