FRACTAL: Advanced Fractional SSM for Long Sequence Analysis
Discover FRACTAL, a novel fractional recurrent architecture enhancing state space models for superior long sequence temporal analysis and prediction accura...
EnvTrustBench: Benchmarking Evidence-Grounding Defects in LLMs
Discover EnvTrustBench, a framework to benchmark and reduce evidence-grounding defects in LLM agents caused by overtrusting environmental evidence.
Preserving Temporal Evidence in Mental Health AI Safety
Explore why preserving temporal evidence is crucial for accurate safety evaluations of mental health AI systems and learn about the SCOPE-MH framework.
Boost RLVR Exploration with Prefix-Tuned Priors
Discover how the IMAX framework enhances exploration in RLVR using prefix-tuned priors, improving reasoning and reducing entropy collapse in AI models.
Can Vision-Language Models Recognize Themselves in Mirrors?
Explore how advanced vision-language models demonstrate self-recognition using mirror tests, revealing new insights into AI self-awareness and cognition.