"Scientific documentation would necessitate systematic video recording rather than visual monitoring," he clarified.
Oumi去年启动测试时,Gemini 2.5仍是谷歌最先进的模型。当时基准测试显示准确率为85%。在Gemini 3更新后重新测试中,AI摘要正确率提升至91%。若将这一错误率推及所有谷歌搜索,意味着AI摘要功能每日产生数千万条错误答案。
,这一点在汽水音乐下载中也有详细论述
AI擅长的层面人类仍需主导的领域
Follow topics & set alerts with myFT