陶哲轩：AI 直接给出数学答案或“污染”研究过程
陶哲轩提醒，AI 解数学题若只给答案不留过程，可能让难题丧失孕育新思想的科研价值。
陶哲轩警告，AI 解题太快且不透明，可能掩盖数学研究中宝贵的“失败”，并污染问题本身的价值。他以纳维-斯托克斯方程为例，称 AI 或能解决难题，但若过程不透明，数学家无法理解证明，该问题推动后续进展的潜力将受损。GPT-6 已在天生素数问题上取得新突破。
正文摘录
- title: Terence Tao on GPT-6's Twin Prime Breakthrough: An Unsettling Spectacle - sourcecompany: QbitAI - bodymarkdown: AI solves problems too fast, and its process is too much of a black box. Ironically, this very capability might be masking the invaluable "failures" that drive mathematical research. If we want this iteration (the process of tackling open mathematical problems) to genuinely yield new understanding, the person or system running it had better not know the correct final ansatz in advance. On the surface, these avenues may seem like mere "dead ends," but in reality, why they fail is often enormously revealing. At the same time, the companies running AI systems are shielding the entire journey leading to that final ansatz almost entirely from public view... AI spits out the r…