gemini-4IMAGE
100 道难题测试：Gemini 4 Argon 答对少但编造最少
同一组 100 道硬题，Gemini 4 Argon 只答对 50 道，却仅编造 7 次、说了 39 次「不知道」，与 Claude Fable 5.1、GPT-6 Astra 形成准确率与诚实度的取舍。
Joel 🌊
@joelsstafford19 天前收录原推发布于 2026.10.17
Joel 🌊 @joelsstafford原推发布于 2026.10.17
The 3 top AIs got the same 100 hard questions Claude Fable 5.1: 67 right, 24 confidently made up GPT-6 Astra: 63 right, 19 made up Gemini 4 Argon: only 50 right, but just 7 made up. It said "I don't know" 39 times Would you rather use the AI that's right more often, or the one that admits when it doesn't know?
x.com/joelsstafford/status/2111381650214990082采集于 2026.10.18 10:08