3qk7 · AI LOCAL LABS
TECHNICAL SCOUTING
AI FRONTIER
Signals worth understanding before they become obvious.
2026-09-29
01
watchagentscapability unlock
Advancing Private AI Compute with secure, server-side memory
Introducing private, server-side memory to Private AI Compute for personal AI.
关键变化是：过去很难落地的能力，正在接近真正可用于产品。
02
watchagentsnew architecture
Cartograph: Federated Tool Discovery with Operator-Attested Retrieval for AI Agents
论文摘要：Cartograph：提出或使用新的基准评测，用来衡量模型在特定任务上的表现。
值得关注的不是又一次参数升级，而是一种可能改变产品设计方式的新架构。
03
testresearchincremental
Auditing and Repairing LLM-as-Judge Failures in a Production Text-to-SQL Pipeline
论文摘要：Auditing and Repairing LLM-as-Judge Failures in a Production Text-to-SQL Pipeline：围绕大语言模型的能力、评测或对齐问题展开，适合关注方法与实验结果。
成熟度已经接近可以实际测试，而不只是继续读论文或看演示。
04
watchagentsnew architecture
Stealth Apart, Harm Together: Skill Cascading Attacks on Skill-Based Agent Systems
论文摘要：A skill is a modular package of natural-language instructions, executable scripts, and reference resources that an agent can load at runtime to extend its capabilities for a s...
值得关注的不是又一次参数升级，而是一种可能改变产品设计方式的新架构。