University of Southern Denmark (SDU) · 5 records

dateidtitleauthorsvenue↑
2026-08-172608.13517DFM Mimir v1: An Open HRM Delivering Frontier Performance at 1B Parameters Using Only Permissible Post-Training DataPeter Schneider-Kamp et al.arXiv29
2026-06-152606.10747The Arbiter Agent: Continually Monitoring Multi-Agent Conversations to Detect Emergent MisalignmentFilippo Tonini et al.arXiv11
2026-06-102606.09707BrainSurgery: Reproducible and Reliable Declarative Weight Manipulations for Model Editing and UpcyclingGianluca Barmina et al.arXiv7
2026-06-102606.09697PsychoSafe: Eliciting Psychologically-Informed Refusals in Large Language ModelsGianluca Barmina et al.arXiv6
2026-06-052606.06286LLMs Can Leak Training Data But Do They Want To? A Propensity-Aware Evaluation of Memorization in LLMsGianluca Barmina et al.arXiv7