AI RESEARCH

Self-Policy Distillation via Capability-Selective Subspace Projection

arXiv CS.CL

ArXi:2605.22675v1 Announce Type: new Self-distillation bootstraps large language models (LLMs) by