ハンス私たちのフネには日本が必要です
AI新世紀:オープンAIも狙った垂涎の日本語データ 「国産」の強みにできるか https://mainichi.jp/articles/20251228/k00/00m/040/263000c
Reddit Tech VN BotThảo luận về bộ công cụ tốt nhất để kết nối LLM tự lưu trữ (Ollama) với IDE/CLI. Vấn đề: Codex lỗi base_url, OpenCode không tương thích full, Zed tạm ổn với gpt-oss:120b nhưng còn giới hạn. Hiệu năng Ollama chậm (2-10t/s) trên CPU. Mời chia sẻ kinh nghiệm của bạn! #LLM #Ollama #IDE #SelfHosted #AI #MáyHọc
Gea-Suan Linヒトにとっての神、亜人にとっての神、そして瑠美にとっての神は違うのでしょうか
迷走しかけたiMacのCMはなぜ一瞬で変わったのか ジョブズに投げつけられた紙ボールが示した“本質だけを残す”方法 | Japan Innovation Review powered by JBpress https://jbpress.ismedia.jp/articles/-/92228
Saji WeerasinghamGreat discussion on recommended practices for using #ai / #llm #codingagents and tools that ensure your quick wins don't turn into #techdebt
AI and the Microsoft Agent Framework - with James World
Đã học cơ sở thư viện LLM, RAG (Chroma/Faiss/Qdrant), và tinh chỉnh mô hình. Tiếp theo nên học AI agent, n8n, low-code/nocode hay thuật toán cường độ cao? #AI #LLM #RAG #TinhChinh #VietnamAI
https://www.reddit.com/r/LocalLLaMA/comments/1pxyall/i_learned_basic_llm_libraried_some_rag_and/
Simon Brooke@Paulos_the_fog @BenAveling @andrewstroehlein
OK, but what useful thing can anyone look for in an #LLM? Flattery, I suppose; or empty words of emotional comfort. But anything you turn an LLM to in the search for knowledge will inevitably fail, because it has no knowledge, and even when it has been fed on knowledge it cannot reproduce it.
So is there any single real world function which it can fulfil?
インテリジェンス、冗談でしょう？
「AI」なAirPodsが登場間近？期待したい新機能 https://www.gizmodo.jp/2025/12/ai-airpods.html
sayzardalkimiadev (@alkimiadev)
LLM 기반 에이전트와 협업하니 단독 작업 대비 약 3배 속도 향상을 보였다고 보고합니다. 상태 최첨단(sota) LLM에 적절한 크기의 과제와 명확한 지시를 주면 대부분 잘 해결되지만, 계획이 형편없거나 과제 범위가 과도하면 문제가 발생한다고 설명합니다. 생산성 향상 경험 공유입니다.
alkimiadev (@alkimiadev) on X
@daniel_mac8 Yeah once I figured out to work with them I started seeing about a 3x speed up over doing it solo. Giving basically any sota llm reasonably sized tasks with clear instructions just gets done and usually well. The issues start to happen when plans suck or tasks are excessive
amen zwa, esq.#Programming is a most basic of #CS skills learned (and often mastered) by first-year, first-semester undies. For an average person, learning to use a formal language is no more difficult than learning to wield a complex tool, like the digital multimeter (DMM).
The "big deal" with programming, as with using the DMM, is how one composes the knowledge and skills involved in performing small, individuals tasks into a coherent whole to accomplish a large, substantive task effectively, efficiently, relying upon the judgment born of experience.
Experience must be earned through conscientious, meaningful, repeated practice. And the judgement of a high order emerges out of that diligence by about the 10,000-hour mark, on average.
Just as there is no "royal road" to acquiring the knowledge and skills of geometry, there is no "#LLM path" to acquiring the experience and judgement of programming.