KK Blog

Menu

Archives

2026 / 07

"AI for Teachers Has a Measurement Problem: Building the Learning-Outcome Feedback Loop"

2026-07-19

洞察 #AI for Teachers#Learning Outcomes#Education AI#ChatGPT for Teachers#Claude for Teachers#Causal Measurement
Teacher AI must connect time saved to teaching practice, student behavior, durable learning, and evidence strong enough to guide school decisions.
"Detecting Browser Agents After Fingerprints Fail: Behavior, Signed Identity, and False Positives"

2026-07-19

调研 #Browser Agents#Bot Detection#Cloudflare Precursor#FP-Agent#Web Bot Auth#AI Security
Browser agent detection now needs three layers: behavioral risk signals, signed identity, and explicit authorization for high-risk actions.
一次 AI 越狱攻击有多严重：CJS、JEF 与 CVSS 评分框架解析

2026-07-18

调研 #AI安全#Jailbreak#CJS#JEF#CVSS#Anthropic#红队测试#LLM安全
越狱成功率无法说明真实危害。本文解析 Anthropic CJS 四轴评分，并对照 JEF 与 CVSS，建立从攻击成功到组织风险的分层评估方法。
How Severe Is an AI Jailbreak? CJS, JEF, and CVSS Scoring Explained

2026-07-18

调研 #AI Security#Jailbreak#CJS#JEF#CVSS#Anthropic#Red Teaming#LLM Security
Anthropic's CJS scores AI jailbreak severity across four axes. This guide compares CJS, JEF, and CVSS, then separates severity from risk.
Claude 内部的隐藏工作区：J-space 如何承载没有说出口的思考

2026-07-18

解读 #Anthropic#Claude#J-space#Jacobian Lens#可解释性#AI安全#全局工作区
Anthropic 在 Claude 内部发现了一个低容量、可语言化且具有因果作用的 J-space。它能承载没有输出的中间概念，却不能证明 Claude 拥有主观意识。
Claude's Hidden Workspace: How J-Space Carries Thoughts It Never Says

2026-07-18

解读 #Anthropic#Claude#J-space#Jacobian Lens#LLM Interpretability#AI Safety#Global Workspace
Anthropic found a small, causally active J-space inside Claude. It carries silent concepts but does not prove subjective consciousness.

2026 / 06

"MRC 协议：OpenAI 新网络层如何支撑大规模 AI 训练"

2026-06-08

调研 #MRC#OpenAI#AI基础设施#网络协议#分布式训练#OCP#RDMA#以太网#超级计算机
2026 年，训练一个前沿 AI 模型需要数万个 GPU 协同工作数月。但制约训练速度的关键因素往往不是 GPU 算力，而是连接这些 GPU 的网络。当一条链路拥塞或一台交换机故障就能让整个训练任务停滞、导致数千块昂贵加速器空转时，网络就成为 AI 基础设施的决定性瓶颈。 OpenAI 的回应是 M
"MRC Protocol: How OpenAI's New Networking Layer Powers AI Training at Scale"

2026-06-08

调研 #MRC#OpenAI#AI Infrastructure#Networking#Distributed Training#OCP#RDMA#Ethernet#Supercomputer
Training a frontier AI model in 2026 requires tens of thousands of GPUs working in tight synchronization for months. Yet the factor that most often li
"B2B Signals 解读：2026年前沿企业的 AI 采用正在拉开不可逆的差距"

2026-06-06

解读 #OpenAI#B2B Signals#企业AI#AI采用#前沿企业#Agentic AI#Codex#AI规模化#组织转型
"OpenAI B2B Signals 报告显示前沿企业每个员工的 AI 智能消耗是典型企业的 3.5 倍。36/64 的体积与深度分裂揭示了为什么大多数企业在 AI 采用上问错了问题。"
"B2B Signals: How Frontier Enterprises Are Scaling AI Adoption in 2026"

2026-06-06

解读 #OpenAI#B2B Signals#Enterprise AI#AI Adoption#Frontier Firms#Agentic AI#Codex#AI Scaling#Organizational Transformation
"OpenAI's B2B Signals report reveals frontier firms use 3.5x more AI per worker than typical firms. The 36/64 split between volume and depth reveals w
Previous Next