KI-Forschungspapiere Täglich

Multimodal Needle in a Haystack: Benchmarking Long-Context Capability of Multimodal Large Language Models

Long Code Arena: a Set of Benchmarks for Long-Context Code Models

Probabilistic Conceptual Explainers: Trustworthy Conceptual Explanations for Vision Foundation Models

Self-MoE: Towards Compositional Large Language Models with Self-Specialized Experts

Breaking Boundaries: Investigating the Effects of Model Editing on Cross-linguistic Performance

Interface Design for Self-Supervised Speech Models

Measuring memorization in RLHF for code completion

Intrinsic Evaluation of Unlearning Using Parametric Knowledge Traces