本期四则近期公开观点，先看 AI 的能力为何会「尖峰化」，再看前沿安全研发的资金问题、编排模型进入网络安全评测，以及用户数据训练政策的透明度。
01｜能力尖峰，不是能力底线
AI competence has always been very spiky, superhuman in some narrow domains and largely useless in others. The fundamental marketing trick of the AI industry is to make you believe the tallest spike is a floor. 1
François Chollet 把「能力不均匀」说得很直白：看到某个任务上的惊人表现时，还要问它能否迁移到相邻任务，以及那是不是稳定能力。
来源：François Chollet · @fchollet · X · 2026 年 7 月 20 日 12:11
콘텐츠 카드를 불러오는 중…
02｜安全研发，长期缺钱
One important strategic opportunity put forward for the EU is pooling strengths with trusted partners facing similar constraints such as the UK, Canada, Japan, Korea, Australia, New Zealand and India. Such an international coalition should prioritize reliable, safe, and secure frontier AI R&D, which remain a persistently underfunded research area. 2
Yoshua Bengio 提到的不是单个实验室的安全路线，而是欧盟与可信伙伴的协作选项：把前沿安全研发当作需要共同补足的长期研究投入。
来源：Yoshua Bengio · @Yoshua_Bengio · X · 2026 年 7 月 21 日 03:19
콘텐츠 카드를 불러오는 중…
03｜智能体，开始测真实安全
Incredibly proud of the Sakana AI team. We have developed an orchestration model right here out of Japan that achieves state-of-the-art performance on real-world cybersecurity benchmarks! 3
Sakana AI CEO hardmaru 称，团队在日本开发的编排模型，已经在真实世界网络安全基准上取得 state-of-the-art 表现。这条帖没有展开模型名称和基准细节，原帖是目前可核对的成果入口。
来源：hardmaru · @hardmaru · X · 2026 年 7 月 21 日 10:15
콘텐츠 카드를 불러오는 중…
04｜先说清楚，数据怎么被训练
Right now it feels like the single biggest competitive advantage an AI lab could have is making it abundantly clear whether and how they will train models on your data. I pay pretty close attention to this and I couldn't confidently summarize the policies for ANY of the lead labs. 4
Simon Willison 把数据训练政策放回产品竞争里讨论：用户需要的不是一句「重视隐私」，而是能准确回答自己的数据会不会、怎样被用于训练。
来源：Simon Willison · @simonw · X · 2026 年 7 月 17 日 22:57
콘텐츠 카드를 불러오는 중…
本期索引：François Chollet · Yoshua Bengio · hardmaru · Simon Willison
댓글
로그인하면 댓글을 작성할 수 있습니다.