gpt-6POST
有人问 Astra 最高推理档的数学输出有没有硬伤
一位研究者称本月核查或形式化了数十个论证，没在 GPT-6 Astra max reasoning 输出里找到关键数学错误。
Lénaïc Chizat
@LenaicChizat29 天前收录原推发布于 2026.10.05
Lénaïc Chizat @LenaicChizat原推发布于 2026.10.05
Relatedly, I’m curious if anybody has found any critical math mistake in the output of GPT-6 Astra max reasoning? I have checked or formalized dozens of arguments this month and not found any. I wonder if formalization will be needed when reliability will be >99.99%.
这条是引用着另一条写的：x.com/scottnarmstrong/status/2106855895493685385
x.com/lenaicchizat/status/2106920528489624043采集于 2026.10.06 10:05