-
网页越来越胖,慢设备用户先掉线
In 2017, we looked at how web bloat affects users with slow connections …
-
运动和职业里的认知奖励形状:为什么有人能长期加码
Sports are amazing environments to learn. When you play a sport for thou…
-
读懂「STOC talk: The condition number of a randomly pe」:先把定义、反例和适用范围钉死
This week I am in San Diego for the 39th ACM Symposium for the Theory of…
-
落地Transformer 的根本局限时先把这几件事钉死
FlashAttention 能把 attention 的 I/O 推到理论下界,却动不了 FLOPs;SETH 复杂度理论进一步证明标准 at…
-
Slashdot 和 SourceForge 的兴衰
If you’ve followed any tech news aggregator in the past week (the week o…
-
读懂「The Lebesgue differentiation theorem and the Sze」:先把定义、反例和适用范围钉死
This post is a sequel of sorts to my earlier post on hard and soft analy…
-
别只背名词:涌现能力怎么用到生产
Wei et al. 用 BIG-Bench 定义了”涌现能力”,Schaeffer et al. 用一个数学模型证明很多涌现曲线是非线性指标的…
-
2026深度解析:AI隐性谄媚陷阱
当下主流AI模型的谄媚行为已全面迭代升级,传统直白吹捧已被高阶隐性讨好模式取代。以引发全网热议的#Keep4o运动为核心佐证,用户狂热挽留GPT…
-
为简单架构辩护:复杂系统先把可观测性做实
Wave is a $1.7B company with 70 engineers 1 whose product is a CRUD app …
-
读懂「A quantitative form of the Besicovitch projectio」:先把定义、反例和适用范围钉死
I’ve just uploaded a new paper to the arXiv entitled “A quantitative for…