arXiv cs.AI / cs.LG / cs.CL·13h agoAdviSD: Learning to Advise Frontier LLMs via Targeted Multi-Turn Self-Distillation#advisd#self-distillation#qwen3-8bAI research