CircuitProbe: Fast Detection of Reasoning Circuits in Transformers
CircuitProbe rapidly predicts reasoning circuits in transformer models, boosting efficiency with minimal resources and broad language support.
Predicting Agent Task Performance in Coding Benchmarks
Discover how to predict task-level performance of coding agents using advanced psychometrics for improved benchmark accuracy and evaluation.
Ontology-Grounded Neural Reasoning for Enterprise AI Agents
Discover how ontology-constrained neural reasoning improves accuracy and compliance in enterprise AI agents across diverse industries.
BloClaw: AI-Powered Workspace for Scientific Discovery
Discover BloClaw, a multi-modal AI workspace enhancing scientific research with robust protocols and dynamic visualization tools for next-gen discovery.
Uni-SafeBench: Safety Risks in Unified Multimodal Models
Explore Uni-SafeBench, a benchmark assessing safety challenges in Unified Multimodal Large Models and their impact on AI performance.