All tags
Posts tagged with "LLM"
Loop Engineering's Missing Building Block: Why Embedded Systems Are a Safer Playground for AI Agents
Addy Osmani's Loop Engineering describes five building blocks for autonomous AI loops, but misses a critical question: when AI says 'done,' how do you know it's actually done? In AOSP / embedded environments, the answer is deterministic verification, the system itself tells you.
Loop Engineering 的第六塊積木：為什麼嵌入式系統天生更適合 AI Agent
Addy Osmani 的 Loop Engineering 描述了 AI 自動化迴圈的五塊積木，但漏了一個關鍵問題：AI 說「做好了」，你怎麼知道它是真的？在 AOSP / 嵌入式環境裡，答案是 deterministic verification，系統本身會告訴你。
AI 為什麼會一本正經地胡說八道？從底層機制到實際可用的工程策略
大模型的幻覺不是偶發的 bug，而是架構天生自帶的系統性風險。從底層原理拆解成因，談工程解法，以及為什麼 AOSP 開發環境本身就是一台天然的防幻覺兜底機。
Why AI Confidently Makes Things Up: From Root Causes to Practical Engineering Strategies
LLM hallucination is not an occasional bug, it's a systemic risk baked into the architecture. A breakdown of root causes, engineering countermeasures, and why AOSP development workflows are a natural safety net for AI agents.
LoRA Space Navigation: Three Ways to Specialize LLMs Without Fine-Tuning
LoRA parameter space can be programmatically navigated, no gradient descent needed. A survey of 15+ papers across three research schools (generators, routers, mergers) and what they mean for agent systems.
LoRA 空間導航：不用 Fine-Tune 就能特化 LLM 的三條路
LoRA 參數空間可以被程式化導航，不需要 gradient descent。涵蓋三個研究流派（生成、路由、合併）共 15+ 篇論文的調查，以及對 agent 系統的實際意義。