模型在非可验证领域能力提升超出预期
精选理由
Mollick发现，AI在创意推理这类没法打分任务上的进步比想象中大，前沿没那么崎岖。
Ethan Mollick指出，缺乏可验证域（如数学、编程）使模型训练困难，但模型在非可验证域（如创意、策略）的表现持续提升。前沿能力的“崎岖度”低于单纯基于可验证性预期的程度，表明泛化能力增强。这一观察基于对多种推理模型在开放式任务上的表现分析。
原文 · Ethan Mollick
While it is obviously true that not having verifiable domains makes training models in those spaces difficult... it is also true that models are also getting much better at non-verifiable domains. The frontier is jagged, but less so than I'd have expected from verifiability alone
