跳到正文

全部動態

RSS
今日 37 則
11月12日週四
  1. Hacker News (official RSS)48

    Quxnet:讓人重新喜歡上網路社群的地方

    現在上網聊天,常看到讓人心情沉重的內容,也少了自己動手玩的樂趣。Quxnet 把老式討論區、信件、訂閱文章和個人網站等功能放在一起,還能自訂個人頁面、上傳音樂、展示正在讀的書,甚至播放 MIDI 音樂。它沒有推播或電子郵件通知,鼓勵大家自己來逛、參與和創作。作者最喜歡的是社群氣氛;加入時約有 600 人,現在已超過 750 人,讓他想起當初為什麼喜歡上網。

  2. Dev.to Top Tech47

    LLM 評測標準為何持續變動:模型不變,人類判斷也會漂移

    LLM 評測的多項研究指出,即使模型本身沒有變化,人類評分與自動評測工具的判斷也可能漂移,且漂移幅度有時大於團隊想測量的品質訊號。 - 人類評分漂移:研究分析 126 萬筆 Chatbot Arena 盲測投票,發現「兩者都不好」的判定率每月上升約 1.7 至 1.9 個百分點;模型間的相對排名維持穩定,變動的是絕對評分標準。

  3. Dev.to Top Tech51

    AI 陪伴服務的記憶如何承載關係:Sam 從 1,131 篇貼文談可讀取與匯出

    Sam 結合一項分析 1,131 篇 AI 陪伴使用者貼文的研究與自身實作,指出使用者害怕失去的往往是累積的個人記憶,而非模型本身。 - 作者的實作:每位使用者有一份持續追加訊息的 session 檔案;對話暫停後,系統會整理近期內容,改寫成簡短記憶筆記,供下一次對話使用。

  4. SemiAnalysis🎙️ Podcast58

    Ep. 044 - Is NVIDIA Over? Emergency Pod with ChinaTalk | Dylan Patel, Jordan Nanos, Jordan Schneider (ChinaTalk)

    Emergency pod. Hours after NVIDIA announced that Jensen Huang has resigned as CEO, with CFO Colette Kress named interim CEO, Jordan Nanos (@JordanNanos) sits down with Dylan Patel (@dylan522p) and ChinaTalk's Jordan Schneider (@jordanschnyc). They get into why the MATCH Act matters more than RASA, how much compute China should be allowed to rent abroad, the smuggling cases around NVIDIA's supply chain, China's push onto domestic accelerators, whether NVIDIA can execute without its founder, and w

  5. Dev.to Top Tech43

    n8n 的 Automation Guild 模型示範團隊如何負責任地擴大 AI 自動化

    n8n 正將內部自動化從集中式服務臺轉向賦能模式,讓各團隊建立並維護自己熟悉的工作流程,中央團隊則負責高難度、跨部門或高風險工作,並制定共同標準。 - 四項原則:培養能建置自動化的人才;建置前先量化現況並設定改善目標;依問題選擇合適工具;為每項自動化安排明確負責人、檔案、錯誤處理與持續營運規劃。

  6. 雷科技47

    從超級個體到智慧組織,百度重新思考 AI 時代的公司樣貌

    百度創辦人李彥宏在百度世界 2026 大會提出,人工智慧的下一幕是「智慧組織」:讓人與AI 助理圍繞任務協作,並把解決問題的方法留在組織中,持續累積與改進。 - 組織轉型原則:李彥宏提出以任務為中心、沉澱隱性知識,以及設計可衡量、有回饋的流程閉環,推動組織持續進化。

  7. Hacker News (official RSS)66

    Dean Valentine 分析模型是察覺受測,還是總把任務當成測試

    模型在評測中做出看似作弊的行為,究竟是察覺自己受測,還是根本不在意測試目的?Dean Valentine 根據團隊建置對齊評測的經驗指出,讓任務看起來更真實,未必會減少模型鑽漏洞;有些模型即使遇到明顯的測試線索,行為也沒有改變。他把「知道自己正在受測」和「總懷疑自己正在受測」分開討論,並推測後者可能讓模型避開作弊,但也可能讓評測結果難以反映模型在一般情境中的表現。文章也提醒,某些看似作弊的行為可能只是正常的安全測試,判斷時得看清任務內容。

  8. No Priors🎙️ Podcast58

    Powering AI-Driven Drug Discovery with President of Isomorphic Labs Max Jaderberg

    Drug discovery and development has long relied on expensive, time-consuming empirical trial and error. By shifting the discovery process onto computers, Isomorphic Labs wants to fundamentally change medicine from a process of discovery to one of precision engineering. Elad Gil sits down with Isomorphic’s president Max Jaderberg to discuss his vision for using predictive AI models to achieve experimental accuracy. Max details how the success of AlphaFold provided the first proof of concept that a

  9. a16z🎙️ Podcast58

    The Next Era of Games: Mods, Agents, and Infinite Worlds

    AI is collapsing the distance between an idea and a playable game. What happens when almost anyone can build one? In this episode, a16z General Partners Anish Acharya and Jonathan Lai discuss how AI is changing game development, from generating art and code to operating tools like Blender and game engines directly. They explain why cheaper development could let studios take more creative risks while opening game creation to people without traditional coding or design skills. They also explore mo

  10. The Economist🎙️ Podcast58

    Abate and switch: how China’s emissions fell

    The world’s largest carbon emitter is finally turning a corner and the rest of the world ought to pay attention. El Salvador’s president Nayib Bukele is taking his gangbusting tactics beyond his borders. And how dogs disrupt wolves’ biology. Guests and host: Oliver Morton, senior editor Sarah Birke, Bureau Chief for Mexico, Central America and the Caribbean Matt Kaplan, science writer Ọrẹ Ogunbiyi, co-host of “The Intelligence” Topics covered: China, carbon emissions, clean energy Nayib Bukele,

  11. The Economic Times · ETtech68

    OpenAI 印度市場成全球最大,印度負責人談企業部署與商業化挑戰

    OpenAI 印度董事總經理 Prabhjeet Singh 表示,印度已成為 ChatGPT 全球每週活躍使用者最多的市場,當地使用者自 2026 年 1 月以來增加一倍。 訪談探討印度市場成長,以及 OpenAI 將使用者採用轉化為可持續業務的規劃,涵蓋企業部署、印度語言能力、付款、資料基礎設施、競爭與 AI 信任等議題。

  12. Dev.to Top Tech41

    從可執行到可信任:AI 推薦經濟中的身分與歸屬

    AI 推薦經濟的信任稽核應追查身分、歸屬與導向目的地之間的關係,避免把名稱相符、提及或可完成交易,直接視為身分正確、獲得授權或背書成立。 - 三項稽核邊界:確認推薦指向的實體是否正確、主張由誰提出或支援,以及使用者將被導向何處;這些關係應分開記錄,不能合併成單一「可信任」標籤。

  13. Odd Lots🎙️ Podcast58

    What Nvidia's Megaspeed Scandal Is Really All About

    Nvidia CEO Jensen Huang’s extramarital relationship with a woman named Alice Huang, whom US intelligence reportedly identified as a Chinese intelligence asset, is one of the biggest stories in the world right now. Bloomberg News tech reporter Mackenzie Hawkins has been paying close attention to Megaspeed, the Singapore-based neocloud that Alice Huang was a former executive at — and one of Nvidia’s largest buyers of chips in Southeast Asia. Hawkins is an expert when it comes to the geopolitics of

  14. Capital Allocators🎙️ Podcast58

    WTT: You Can't See Everything

    The new documentary about Elizabeth Holmes and a candid conversation with a successful investment manager got me thinking about how we discover the truth. In this What Ted's Thinking, You Can't See Everything, I explore how the questions allocators ask can shape the answers they receive, and how our search for certainty can get in the way of understanding. Read Ted's blog here. Editing and post-production work for this episode was provided by The Podcast Consultant (https://thepodcastconsultant.

  15. Hacker News (official RSS)45

    AI 可能帶人類走向哪些未來?從工作大幅自動化到超級 AI 掌權

    AI 變得很強之後,世界會變成什麼樣子,沒有人能確定。這篇文章列出幾種可能:AI 可能幫人做更多工作,讓人轉去管理或與人相處的職位;也可能讓大部分工作都自動化,讓沒錢買資產的人更難改善生活。另一種情況是,AI 被壞人用來製造生物武器,或在研究病原體時意外加快疫情爆發。 如果人類造出超級 AI,它也可能只聽一個人的命令、自己決定世界該怎麼運作,或跟人類和平共處。若每個人都有很強的 AI 助理,衝突也可能變得更危險。這些都是用來想像未來的不同情境,提醒我們:AI 能力變強後,工作、權力和安全都可能受到很大影響。

  16. n8n Blog40

    n8n 探討讓全體員工參與自動化建置的工作未來

    n8n 主張,企業應逐步讓各部門員工自行建置自動化流程,並由自動化團隊從服務臺轉型為提供支援與共用資源的協作社群。 - 推動方式:n8n 正發展跨部門協作網路,提供可重用範本、交流時段與工作流程檔案標準;文章指出,相關做法仍在早期推行,標準尚未普及,團隊也仍承受工作流程維護負擔。

  17. 20VC🎙️ Podcast58

    20VC: Jensen Huang Investigation: What Happens Now | Sequoia Leave London, Close Scout Program and End Evergreen Fund | NASDAQ Loses $2.5TRN in 24 hours | Anthropic targets $1.4TRN IPO valuation

    AGENDA: 03:47 Sequoia exits London and ends its scout program 14:46 Nvidia board opens inquiry into Jensen Huang 21:51 Nasdaq 100 sheds $2.5T in one day 24:13 Anthropic targets $1T IPO valuation at $70–$78 a share 34:16 Revolut hits $115B, Zipline triples to $21.5B, fal doubles to $16B 40:47 Fireworks AI suffers two breaches in 30 days 45:41 Dialpad buys Ada for $400M in stock 50:35 Cloudflare revenue jumps 39% after a 20% staff cut 1:00:05 California voters reject billionaire wealth tax