AI
SkillEffect: Checked Lowering for Memory-Bounded Agent Tools
Researchers have developed a system called SkillEffect to help language models manage memory usage when interacting with tools. The system uses a chec...
AI
AI
Memory Is Communication: The Frontier Between Remembering and Signaling
Researchers propose a new framework to understand how agents allocate information between their own memory and communication with peers. They introduc...
AI
DiSCO: Defending text-to-image generation through distribution-guided contrastive prompt optimization
Researchers have proposed a new defense mechanism called DiSCO to prevent text-to-image generative models from producing harmful content. The system o...
AI
KernelArc: A Multi-Agent Framework for GPU Kernel Optimization
Researchers have developed KernelArc, a framework for optimizing GPU kernels across different workloads. The system uses multiple agents that run in p...
AI
A decodability criterion predicts when hidden-state selection beats majority voting in large language models
Researchers have proposed a method called CASE to improve the accuracy of large language models by selecting the best answer based on hidden-state sig...
AI
Toward Personal Intelligence Through Cooperative Observation
Researchers propose a framework called 'cooperative observation' to improve the performance of personal AI systems. They argue that these systems need...
AI
KnowSim: Evaluating Information Calibration in LLM Assistants with User Simulators that Learn
Researchers have developed a new framework called KNOWSIM for evaluating the performance of Large Language Models (LLMs) in collaborating with users o...
AI
Synthesizing Feature Extractors: An Agentic Approach for Algorithm Selection
Researchers have developed an automated approach to designing feature extractors for constraint satisfaction problems. This method uses large language...
AI
Benchmarking the Benchmarks: Evaluating Automated Safety Benchmarks for Small Language Models
Researchers evaluated five widely used benchmark suites on 26 open-source small language models to determine their effectiveness in assessing safety. ...
AI
Fool's Gold: Defensive Deception Against Safety-Removal Attacks on Open-Weight Models
Researchers have developed a new defense mechanism called 'Fool's Gold' to protect open-weight language models from safety-removal attacks. These atta...
AI
Explicit State Elicitation Is Not Enough: A Controlled Audit of Memory-Policy Classification
Researchers have developed an audit protocol to test the performance of personalized agents in decision-making tasks. They created a synthetic develop...
AI
Do LLMs Know a Good Hypothesis When They See One? Logit-Based Energy Scoring Outperforms Prompted LLM-as-Judge for Scientific Hypothesis Ranking
Researchers proposed a new method to evaluate scientific hypotheses generated by large language models (LLMs). Instead of relying on LLMs as judges or...