[논문리뷰] Sharpening Tax in Post-Training본 논문은 Large Language Models (LLMs)의 Reinforcement Learning (RL) Post-training이 Solution Coverage를 희생하면서 Sampling Efficiency를 향상시키는 'Accuracy-Coverage Tension' 문제를 다룹니다.#Review#Reinforcement Learning (RL)#Large Language Models (LLMs)#Post-training#Sharpening Tax#Solution Coverage#Sampling Efficiency#Agentic Tasks#Posterior-Tempered Group Sampling (PTGS)2026년 10월 1일댓글 수 로딩 중