AI 日報

聯邦披露檔案顯示，Anthropic、OpenAI 與輝達在今年 7 月至 9 月均錄得各自最高的單季遊說支出，其中輝達以 220 萬美元居首，Anthropic 為 210 萬美元，OpenAI 為 150 萬美元。這一數字雖與微軟、谷歌等長期深耕華盛頓的科技巨頭相比並不突出，卻清楚表明這幾家 AI 公司已意識到塑造政策的重要性。

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：Today, I’m talking to Bill Gates, cofounder and former CEO of Microsoft and chairman of the Gates Foundation. You may have noticed Bill in the news more lately — he’s been on something of a tour for the past few months, ever since publishing an essay in late August titled “The turbulent AI era is here. The choices we make now are critical.” In his piece, Gates ratcheted up his rhetoric around AI safety, writing that “even under the best circumstances, the transition to this new AI era will be one of the most turbulent times in human history.” In interviews since, Gates has called AI “the most dangerous thing that humans have ever gone near,” and that AI “makes nuclear weapons look like nothing.” It is clear he does not think the current policy response from the…

評論員Chris Stokel-Walker指出，荷蘭政府網站（含數字身份登入服務）、芬蘭的公共服務以及維基媒體旗下部分系統相繼下線，但這些大多並非網路攻擊所致，而是運營方因繼續線上風險過高而主動關閉。他認為，僅靠少數疲憊的開發者已難以為繼，AI時代的數字基礎設施正暴露出深層的脆弱性。

IEEE Spectrum 講述了聖安東尼奧軟體工程師 Nikhil Jathar 透過 IEEE-USA 組織的國會訪問日兩次走進華盛頓遊說的經歷。文章說明工程師個人倡導與受《遊說披露法》約束的專業遊說之間的區別，並借眾議院科學、空間與技術委員會工作人員之口指出：準備充分、訴求具體、帶著真實專業知識和個人信念的工程師，更容易贏得國會工作人員的信任，從而影響科技政策。

包括輝達、微軟和思科在內的AI基礎設施企業團體向英國政府提交意見書，警告部長們只有約12個月時間化解公眾對資料中心的反對，否則可能面臨類似美國的抗議，進而威脅專案開工。連續乾旱的夏季和不斷上漲的能源賬單正加劇公眾對大型AI計算設施規劃的審視。

Joss Moorkens 與 Monja Jannet 撰寫的一篇書章，為翻譯領域引入資料倫理議題，討論版權、資料所有權與偏見等普遍性問題，以及少數語言社群和數字鴻溝兩側群體所面臨的特殊倫理挑戰，並探討責任、監管以及翻譯研究界對資料倫理日益增長的關注。

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：arXiv:2610.28716v1 Announce Type: new Abstract: Effective decision-making in human-centric environments such as Intelligent Tutoring Systems (ITSs) and clinical healthcare requires policies that are adaptive, data-efficient, and ethically deployable. Apprenticeship Learning (AL) infers policies directly from expert demonstrations by recovering the underlying reward structure, enabling strictly offline learning without environment interaction. This dissertation develops a progression of offline AL frameworks organized around two challenges: reward modeling and demonstration quality. For reward modeling, EM-EDM clusters demonstrations by latent reward functions and induces cluster-specific policies, and THEMES partitions trajectories into temporally coherent sub-…

OpenAI的模型現已登陸全部三大主要雲平臺，這標誌著其企業戰略的進一步調整。

輝達推出 AIOps 平臺 Minerva，把基礎設施與 AI 工作負載的遙測資料匯聚起來，並藉助專用 AI 模型把海量訊號轉化為可執行的洞察，幫助團隊更早發現問題、提升 GPU 利用率與 AI 工廠產出。IREN 等多家合作伙伴計劃部署，戴爾、聯想、Supermicro 將把它納入其 AI 工廠運營產品組合。

Moshi 創始人兼 CEO Anekha Sokhal 在 AI Engineer New York 與 swyx 對談，講述她從數學、摩根大通能源衍生品交易員到創業者的歷程，以及 Moshi 如何把海量市場資料轉化為與交易員倉位相關的即時洞察。她形容 Moshi 像是 Bloomberg、Claude 和 Netflix 的結合體，並談及可信資料來源、文件抽取、自建還是採購模型，以及當下 AI 工程師被要求“你就是整支團隊”所承受的壓力。

Moshi 執行長 Anekha Sokhal 在 AI Engineer New York 與 swyx 對談，講述她從數學背景和摩根大通能源衍生品交易員成長為創始人的經歷，以及 Moshi 如何把海量市場資料轉化為貼合交易員持倉的可操作洞察。她形容 Moshi 是“Bloomberg、Claude 和 Netflix 的結合體”，並談及可信信源、文件提取、自建模型與採購模型的取捨，以及 AI 工程師被要求“你就是團隊”時所承受的壓力。

這篇教程介紹瞭如何僅用約 150 行 Python 程式碼、不依賴框架，構建一支由三個 AI 代理組成的“夜班”團隊：Scout 負責研究、Quill 負責撰寫、Red 負責編輯；它們靠檔案交接任務、共享同一個 LLM 引擎，並透過程式碼強制的守衛規則和人工審批流程，確保任何對外發布的內容都需經理批准後才上線。

投資研究公司Citrini Research與Anthropic合作，嘗試用尚未釋出的Claude模型從頭設計蠍子抗蛇毒血清。Claude生成大量候選蛋白，其中17個成功與毒素結合，但專家指出這僅是整個藥物研發流程中相對容易的一步，距離真正的藥物還很遙遠，部分興奮為時過早。

Modal 釋出 Spindle，一個開源且相容 Tinker 的訓練伺服器，可執行在使用者自己的 Modal 工作區內。它讓團隊用 GPU 時長而非按 token 計費，並能自定義模型與功能，只需一條命令即可部署。

本文探討 AI 訓練環境如何從單純的評測工具演變為模型能力的核心要素，追溯了從 Atari、OpenAI Gym、AlphaZero 到 MuJoCo，再到支撐 Tulu 3 和 DeepSeek-R1 的 RLVR 式後訓練的演進路徑，並梳理了 SWE-Gym、WebArena、OSWorld、TheAgentCompany、Tau2-bench 和 AgentMercury 等新一代智慧體環境。文章還提出了優質環境應具備的要素：與真實結果一致的獎勵、因果保真度、可負擔且可重複的練習、結構化難度，以及能夠遷移到訓練世界之外的改進。

IBM推出2B引數的Granite Speech 5.0 2B開源語音語言模型，採用Apache 2.0許可，支援6種語言的語音轉文字，並在Open ASR Leaderboard上以5.68%的平均英語WER位列第二。IBM同時更新了邊緣端Turbo CTC和說話人分離模型4.1 2B Plus。

在柏林GTC大會上，多家歐洲銀行、支付公司和金融資料提供商展示了基於NVIDIA加速計算、開放權重模型和自有基礎設施的生產級AI系統。Revolut以PRAGMA基礎模型重構技術棧，Starling Bank、Lloyds、bunq、NatWest、Santander、LSEG、HSBC和Stripe等機構也在主權雲、交易資料建模、代理助手、語音識別和欺詐檢測等場景中取得顯著效率與成本收益。

NVIDIA 在 GTC 柏林大會上宣佈擴充套件 Cosmos 平臺，新增訓練工具、模擬檢查點和 Cosmos Data Factory 藍圖，旨在幫助開發者更快速地構建物理 AI 世界模型。Cosmos 在六個開放模型排行榜上進入前五，新工具覆蓋資料生成、模型訓練最佳化和模擬評估等環節。

NVIDIA在GTC柏林展示了Earth-2開放AI模型家族的最新進展，包括針對歐洲訓練的區域短時預報模型StormScope EU。AXA、Vaisala Xweather、曼徹斯特大學、TotalEnergies和Space42等組織正將這些模型應用於颶風風險分析、閃電資料融合、空氣質量建模、太陽能預報和雷達地形估計等場景。

NVIDIA及其合作伙伴正與歐洲多國政府合作，利用開放模型推動公共部門AI應用，涵蓋歐盟委員會、波蘭、英國、德國、法國和希臘等專案，旨在提升公共服務效率並確保資料主權。

本文是一份可復現的實測教程，用固定隨機種子從 LibriSpeech test-clean 和 test-other 取樣，在 Apple M4 的 CPU 上對比 Cactus Compute 的 16.9 MB 語音識別模型 Whistle 與 Whisper tiny、base。作者採用配對說話人聚類 bootstrap 區間並做 Holm 校正後發現：在 test-other 上 Whistle 顯著優於 Whisper base 的貪心解碼，但與 beam search 解碼相比沒有顯著差距；速度優勢存在，但音訊越長差距越小。教程還測試了 30 秒單次視窗截斷問題、流式延遲、時間戳、關鍵詞偏置、多語言、噪聲、嵌入以及語音呼叫工具。

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：arXiv:2610.28843v1 Announce Type: new Abstract: Visual question answering requires models to assign attributes and relations to the correct objects, a problem known as visual binding. Structured prompts make these assignments explicit through labeled fields. Yet an improvement from the full prompt does not reveal which labels help or how they interact. First, we isolate field-label effects across ten models and six tasks by holding the image and claim fixed. These comparisons reveal differences that overall accuracy can hide. With a relation label present, adding object-role labels helps Qwen3 but hurts Qwen2 on Spatial. Second, we compare auxiliary hints that preserve or swap the question's role assignments against a no-hint baseline. Across two image sets, Qw…

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：arXiv:2610.28829v1 Announce Type: new Abstract: Instruction-guided video generative models offer a scalable solution for stress-testing autonomous driving systems by simulating diverse environmental conditions. However, driving scenes operate within a dense Operational Design Domain (ODD) governed by strict traffic semantics. When applied to global driving-condition edits, general-purpose instruction-guided video editors frequently corrupt these semantics, leading to structural hallucinations and ``temporal shock.'' This work distills and protects traffic semantics during video editing at three levels. First, we construct a hybrid dataset by combining temporally aligned real-world videos with synthetic edits, balancing foundational scene structure with unambigu…

arXiv 上一篇被 NeurIPS 2026 接收的論文提出 Register Anything Model（RAM），這是一種在室內、室外和物體等多領域資料上聯合訓練的統一點雲配準模型，可跨感測器和場景魯棒估計 6-DoF 變換。其核心是幾何感知歸一化策略，將點雲自適應體素化並歸一化到標準幾何空間，再配合由粗到細的分層框架和高效的幾何 Transformer。在 12 個基準上，該方法在內點率和配準召回率上達到最佳水平，甚至優於針對每個單獨領域訓練的模型。

本文提出 JUICE，一個在凍結的影像與文本預訓練表徵空間上執行的單一流匹配 Transformer，兩個模態各有獨立噪聲水平，透過哪些模態保持乾淨、哪些需要生成的設定來切換影像生成、影像描述、視覺問答與聯合生成等任務。在 2.1B 引數、224x224 解析度且不使用語言模型初始化的條件下，它最高達到 GenEval 0.88 與 VQAv2 73.1，並在需要雙模態互相配合的迷宮路徑規劃任務中顯著優於獨立取樣。

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：arXiv:2610.28840v1 Announce Type: new Abstract: Early-grade reading screening is critical for identifying children at risk of reading difficulty. Existing AI-based screeners are often limited in the scope of their analyses and evaluation metrics, or are proprietary systems that lack transparent model specifications and evaluation. We present KIRA, a dual-pathway screening system that pairs word-level fluency scoring with detailed phoneme-level error detection. We adapt pretrained word and phoneme recognizers to children's speech, reducing word error rate from 13.88% to 8.84% and phoneme error rate from 41.66% to 23.00%. Powered by these child-adapted models, we implement Word Reading (WRE) and Oral Reading Fluency (ORF), two foundational tasks in K-2 screening.…

來自 arXiv 的一篇新論文提出了 Lift-Bench 與 Lift-Train，用於評估和訓練語言模型的指令跟隨能力。其核心創新是把約束限定在文件的特定章節內，從而更精細地控制約束密度；基準同時包含可程式驗證的約束和由評審判斷的約束，並同時提供主任務約束和指令跟隨約束，以緩解淺層獎勵駭客問題。作者發現前沿模型在較難的子集上表現吃力，而在 Lift-Train 上訓練比在常見的 IFTrain 上訓練更不易出現獎勵駭客。

該論文研究大型語言模型在處理法律敘事時，其內部啟用中是否存在與CATO全部26個法律要素相對應的表徵。作者在Qwen3-1.7B、Qwen3-8B和Gemma-2 2B三個開放模型上進行線性可解碼性與稀疏自編碼器稀疏定位實驗，發現全部26個要素在三個模型中均可線性解碼，最佳層測試AUC為0.86至1.00，並報告了78個模型特定的層-特徵座標。

該論文指出現有的LLM定價智慧體合謀評估只關注結果，卻忽略了合謀的關鍵——對作弊者的懲罰執行。作者提出一種審計方法，透過在對市場的反事實副本中強制某一成員作弊來檢驗執行能力，並研究強化學習在定價與交易中究竟教會了智慧體什麼。

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：arXiv:2610.28772v1 Announce Type: new Abstract: Psychotherapy fidelity assessment measures how faithfully a therapist adheres to a treatment protocol, a critical determinant of positive patient outcomes. However, fidelity assessment is bottlenecked by manual expert review, rendering the process labor-intensive and unscalable. To scale therapy supervision and support clinician self-learning, we present an integrated application of large language models (LLMs) that automates psychotherapy fidelity workflows and generates actionable supervisory feedback from authentic, long-form clinical transcripts. We evaluate three open-weight models across multiple task decomposition configurations and reasoning strategies, demonstrating that evidence-first reasoning improves…

研究者提出 SkillEval 評估框架，利用大語言模型的元認知能力從題目中自動識別可解釋的技能，並結合心理測量學中的認知診斷模型，為每個模型估算各項技能的掌握水平，從而替代傳統上把所有題目聚合成單一分數的粗粒度評測方式。

一篇 arXiv 論文探討了 Gemma 4 與 EmbeddingGemma 2 共用的 3.05 億引數音訊編碼器在不改動任何權重的前提下，用於沙特與海灣阿拉伯語語音識別的效果。作者構建了兩個系統：僅訓練 CTC 讀取頭的凍結方案，以及在其上增加低秩介面卡、混合 CTC/注意力讀取頭與子詞語言模型聯合解碼的增強方案。增強系統在多個公開測試集上與 20.7 億引數的 Cohere Transcribe Arabic 在詞錯誤率上統計不可區分，部分場景顯著更優，並可在單張 GPU 上達到 139 倍即時速度；但其在長篇沙特語音的詞錯誤率上仍落後，在盲測公共呼叫中心樣本中排名 15 個系統中的第 4。

arXiv 新論文提出運動等變卷積LSTM（MELSTM），將序列資料中的運動視為一種規範對稱性，用速度模型充當規範場，在預測、分類和動作識別任務上優於同等引數量的卷積LSTM與流等變模型。

該論文提出ReAPO，一種從隨機最短路徑（Stochastic Shortest Paths）視角重新審視信用分配、由GRPO簡單修改而來的方法。它把額外生成預算分配給失敗軌跡，使成功的重試得到強化而不獎勵此前的失敗。在Qwen3與OLMo-2（0.6B至4B）以及八個數學與科學推理基準上，ReAPO的首輪嘗試表現優於GRPO和近期基線；在Qwen3-0.6B上，它以少39%和少31%的生成token分別達到GRPO雙倍組大小和雙倍訓練步數的峰值準確率。

這篇論文用經過僅基於驗證集調參的簡單基線與三項近期研究中報告的 GraphLLM 節點分類結果對比，發現這些方法的領先幅度中位數僅為 2.3 至 2.8 個百分點。作者同時記錄了已釋出程式碼和論文中的多項評測風險，包括在測試集上選擇超引數、訓練中使用測試池節點標籤、未披露地複用基線數值，以及 Cora 載入器中的文本與標籤錯位問題。

該論文提出STAR-XPC模型，用集合Transformer將聚合物複合材料的分子組成、質量分數與四項體相物理描述符直接對映為X射線透射比I(t)/I0，推理時間僅1-3毫秒，精度接近蒙特卡洛模擬，可作為輻射遮蔽研究中昂貴模擬的快速替代方案。

來自 arXiv 的會議預印本提出 TF-IG，一種雙分支、互動門控的時頻學習框架，用於基於光電容積脈搏波（PPG）的心率估計。該模型僅有 84.6K 引數，卻在 BIDMC 資料集上取得 1.227 BPM 的平均絕對誤差和 0.9934 的皮爾遜相關係數，展示了在輕量級可穿戴場景下兼顧精度與複雜度的可行路徑。

該論文提出KVLearn，一個面向分離式LLM服務架構的學習型KV快取保留框架，透過預測字首複用機率並結合重計算、傳輸與儲存成本做出保留/准入決策，並線上自適應調整閾值。在文本與多模態負載上，它將首token端到端時延較無快取降低最多56%，較LRU-Pool降低38%，較Mooncake式基線降低33%，同時將節點間KV傳輸量較LRU-Pool減少最多53%。

賓夕法尼亞大學的研究人員用大語言模型分析了近 15 萬條 Reddit 帖子和評論，找出約 1.4 萬名自稱服用過未獲批版本 retatrutide 的使用者。分析顯示，在可醫學歸類的症狀中，食慾增加是最常見的自述症狀，與二期臨床試驗中以噁心等消化道反應為主的結果不同；此外，睡眠障礙、心率升高和肌肉相關症狀的比例也高於此前對司美格魯肽和替爾泊肽的類似分析。研究團隊強調，這些網路自述不能證明藥物導致了上述症狀，而且灰色市場產品的成分和劑量無法核實，該研究的意義在於為後續更嚴格的獨立研究提供假設線索。

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：Recursive has released Flint, an Apache 2.0 local coding model with 40B total and 2.5B active parameters. It reports 74.2% on SWE-bench Verified and 125+ tokens per second on an M4 Pro MacBook Pro. The post Recursive Releases Flint, an Open-Weight 40B MoE Local Coding Model for Agents with 2.5B Active Params appeared first on MarkTechPost.

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：Train Qwen3 on AWS Trainium with torchtitan three ways on a single trn2.48xlarge: the Hugging Face Transformers v5 modeling backend, torchtitan's native Qwen3, and NKI mixture-of-experts kernels. The same torchrun launcher and config structure work unchanged across all three paths, including loading and saving published Hugging Face safetensors.

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯： Today's PHP Security Release, and How We're Strengthening PHP Security If you're running anything on PHP it's vital you upgrade - today's updates (across the entire PHP 8.x line) address two remote code execution holes, and there have been exploits circulating since last night, thanks to a leak in the coordinated disclosure process. This post from the PHP Foundation explain what went so wrong: On Tuesday, October 20th, at around 15:00 UTC, the two CVE IDs began circulating on online forums and social media. They came with a wrong release time and a wrong CVSS score of 10.0 (the actual score is 9.2). No technical details leaked. In response, the PHP project brought its pre-announcement forward and published it at 19:30 UTC. It also moved the release from Octobe…

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：Introducing OCRia-1 Today, we are introducing OCRia-1, the world’s best document parsing model. It is a 4-billion-parameter vision-language model (VLM) that is state-of-the-art in both accuracy, cost and latency. On Par…

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：← Blog Integrated embedding on the Documents API Manish Talreja, Gavin Johnson Oct 22, 2026 Product Share: Today, integrated embedding comes to the Pinecone Documents API, bringing embedding and search together in one A…

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：← Blog When vectors became an implementation detail How Pinecone built integrated embedding into the Documents API Silas Smith Oct 22, 2026 Engineering Share: Nobody searches for a vector. People describe what they're l…

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：Consider this age-old riddle: “What kind of house do you enter blind and leave seeing?” While English speakers may instantly recognize the answer as “a school,” AI models can struggle with the same riddle if asked in an…

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：AI models The top ranked decision models in the world: mammoth-1 and elephant-1 mammoth-1 and elephant-1 are decision models trained by our internal research team, Base Labs. They are open-source and free to try for a w…

Simon Willison 在其部落格中收錄了 Armin Ronacher 的一句諷刺性評論：AI 固然有趣，但過去三十年我們寫出了大量糟糕的軟體，如今宇宙正在為我們的罪過懲罰我們。

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：Did one of your favorite websites go dark on Wednesday? Some did, as operators took services offline while they waited The post AI found a critical PHP flaw. The bigger failure began in 2024. appeared first on The New Stack.

Perplexity 與 NVIDIA 研究人員提出 JOLT（Joint On-policy Learning and Teaching），一種讓單一模型同時扮演教師與學生兩種角色的後訓練方法。教師在訓練中同步更新，使其反饋對學生始終可用。在 MATH 基準上，JOLT 以約 6.5 倍更少的已處理訓練 token 達到與 GRPO 相當的準確率；加入學生自身結果獎勵的 JOLT+ 在全部五個測試場景中平均分最高，其中 AppWorld 上比 GRPO 高 9.9 個百分點。

該論文提出 T-WLA 世界-語言-動作架構，透過共享視覺記憶同時驅動基礎動作生成與輕量級觸覺修正，使機器人在視覺策略更新之前就能對接觸變化做出快速調整。在指關節運動預測任務中，該方法將 MPJPE 降低 30%，並在四項真實機器人靈巧操作任務中取得約 50.0% 的平均成功率。

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：arXiv:2610.28791v1 Announce Type: new Abstract: Target-based extrinsic calibration estimates the rigid transform between sensors from a few observations of a known target, and its result is commonly assessed by the root-mean-square error of the fitted residuals. However, this scalar error is computed on the same observations and therefore cannot indicate whether to accept the result, re-extract, recollect or add observations. This paper presents VeriCalib, which formulates extrinsic calibration as a verifiable estimation problem and reports a calibration state. The key insight is a two-level extrinsic compensation mechanism, in which the extrinsic compensates the rigid part of each observation error at the residual level, and rotation and translation compensate…

該論文提出一種用於物理人機互動（pHRI）的新框架，將魯棒力/力矩（wrench）觀測器與位姿-力混合控制器結合，透過神經網路補償摩擦與未建模動態、利用任務-零空間分解抑制人類擾動，從而在不使用力/力矩感測器的情況下實現位姿與力的可靠跟蹤。

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：arXiv:2610.28701v1 Announce Type: new Abstract: Accurate ground plane normal estimation is essential for odometry in spherical mobile mapping systems (SMMS), which face challenges from rapid rotations and limited ground coverage. We propose two RANSAC-based approaches: single-scan estimation with normal vector scoring and multi-step validation, and LiDAR-inertial odometry (LIO)-based estimation using a global 3D map. Evaluated across diverse real-world datasets against terrestrial laser scanner ground truth, results show that LIO-based estimation significantly outperforms single-scan methods in accuracy and robustness. Integrating ground normal estimates into the motion model improves pose estimation, supporting autonomous navigation in challenging and hazardou…

該論文提出 DAFIC 框架，將單目深度資訊引入小樣本影像分類，透過 RGB-D 自監督預訓練、深度引導幾何自適應模組與空間雙向塊匹配，緩解背景雜亂和 RGB 外觀相似造成的視覺歧義。該工作已被 WACV 2027 接收，並在四個基準上取得優於現有方法的表現。

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：arXiv:2610.28835v1 Announce Type: new Abstract: Conventional visual sensing faces a fundamental dilemma: information that should remain private today may need to be recovered tomorrow. This challenge is becoming increasingly important as cameras become pervasive in our daily surroundings. Existing safeguards either protect visual content only after an image is formed or suppress information during capture in ways that prevent its later recovery. Here, we show through an information-theoretic analysis that privacy and recoverability cannot be achieved simultaneously in static optical systems. Information preserved in a measurement for later reconstruction also remains available to an observer of that measurement, limiting the privacy such systems can provide. To…

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：arXiv:2610.28827v1 Announce Type: new Abstract: We present WorldStream, an open-ended 3D world model with three simultaneous capabilities: it extends playable terrain beyond any pre-specified map, acts on open-ended language in the running world, and restores visited places with their edits. A hierarchical planner over a typed decision model emits structured decision state: Macro fixes settlement, ecology, transport, and hydrological structure across a group of tiles; Micro specializes that contract into tile semantics under measured terrain and accepted neighbor commitments. Deterministic local code compiles accepted plans into terrain, roads, settlements, vegetation, and entities. Streaming separates their reach: plans run ahead along the predicted corridor,…

arXiv 論文 SAIL 提出一個可微框架，將 Moran's I、Geary's C 等經典空間自相關統計量推廣為圖上的可學習運算元：它聯合學習特徵投影與節點間鄰域親和度，輸出可解釋的區域性與全域性空間一致性描述子，既能用於監督預測，也能無監督地發現空間自相關的模式與熱點。

一篇被 WACV 2027 接收的論文對 MS1MV3 人臉識別訓練集進行受控審計，識別出五類殘留資料缺陷，並透過基於 IResNet-50 與 ArcFace 的受控訓練驗證：針對性地修正缺陷可在減少最佳化步數的同時維持甚至提升識別效能，其中重複影像去除的收益最為明確。

一篇新論文研究如何在多語言預訓練中顯式利用平行資料的結構，發現單獨使用平行預訓練（PPT）並不總能提升語言理解能力，但採用“先用PPT構建跨語言基礎、再用標準預訓練進行適配”的兩階段方法，在高資源和低資源場景下都能穩定超過純標準預訓練。

該論文提出 SimHack，一種基於少量小規模參考強化學習執行學習獎勵駭客訓練動態的模擬器，可在實際執行前預測目標訓練配方下可檢測獎勵駭客行為何時成為主導，並據此篩選低風險課程。

該論文提出同步神經拓撲最佳化（SNTO）框架，用定製神經網路同時表示狀態場、伴隨場與設計變數，並在單一最佳化迴圈中聯合訓練，以連續伴隨分析與自動微分實現多百萬自由度應力約束拓撲最佳化的可擴充套件求解。

這篇綜述由Hamza Kheddar撰寫，系統梳理了100多項關於在FPGA和積體電路平臺上實現強化學習的研究，按強化學習家族、抽象層次、硬體實現策略和應用領域進行分類，並比較資源佔用、延遲、吞吐量、功耗與效能增益，同時指出反覆出現的架構模式、實踐侷限與開放挑戰。

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：A rush of new data centers could bring air pollution roaring back to levels not seen since the 2000s in Texas, which has become ground zero for the physical infrastructure behind generative AI. New and proposed data centers are demanding a tremendous amount of electricity. How developers and utilities meet that demand - with gas turbines or renewable energy - will determine just how dirty things will get. So far, though, things are looking grim. Tech companies and data center developers are devising new ways to get enough electricity, including building out their own power sources. But if all planned data center projects were to plug into … Read the full story at The Verge.

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：PHP powers roughly 70% of the world’s websites, according to Wiz Research, so a flaw in its core engine affects The post AI found flaws in the code behind 70% of the web. The fix took two months. appeared first on The New Stack.

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：U.S. Hunger, a nonprofit addressing food insecurity, uses Amazon Quick to connect organizational data with leadership decisions. Leaders explore applicant data directly, share KPI definitions and financial context in Quick Spaces, and investigate policy questions with Quick Research.

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：For decades, a fundamental question about the difficulty of computing in high-dimensional spaces has resisted proof. New research resolves this question, removing a major theoretical obstacle to understanding the efficiency of randomized algorithms used across science and computing. The Kannan–Lovász–Simonovits (KLS) conjecture was proposed in 1995. It asks whether, for a broad class of high-dimensional […] The post Solving a 31-year-old grand challenge in mathematics – the KLS conjecture appeared first on Microsoft Research.

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：Dr. Christopher Stubbs, an experimental physicist and former Dean of Science at Harvard University, has joined Anthropic’s Long-Term Benefit Trust. He brings to the Trust decades of scientific expertise across observati…

Monish 在 Product Hunt 上釋出，主打讓 AI 智慧體直接顯示在 iPhone 鎖屏上，使使用者無需解鎖手機即可檢視智慧體狀態與動態。

Anekha Sokhal 曾是能源衍生品交易員，如今創辦 Moshi，為商品與能源交易員打造一款推薦引擎，把海量市場資料轉化為與個人持倉相關的即時洞察。她在 AI Engineer New York 與 swyx 對談，講述從數學和摩根大通到創業的歷程，並將 Moshi 形容為“彭博、Claude 與 Netflix 的合體”，強調有用的 AI 必須貼合人們實際的工作方式。

谷歌在 Linux 基金會 AGNTCon+MCPCon 大會上宣佈，將內部支撐 Gemini 智慧體的 Agent Compute Engine 於今年晚些時候向 Google Cloud 使用者開放預覽，為每個 AI 智慧體提供硬體隔離的 microVM，兼顧容器速度與沙箱安全。谷歌還提供開源專案 Agent Substrate，供希望在自有 Kubernetes 環境中執行智慧體的團隊使用。

1Password 推出“1Password for Agents”功能，允許 AI 智慧體在不檢視或接觸使用者密碼的情況下登入賬戶。智慧體透過唯一數字金鑰向 1Password 驗證身份，使用者在本人裝置上批准登入請求後，由 1Password 代為填入憑據，流程類似雙因素認證。該功能將支援 Meta Muse、ChatGPT Work、Claude Duet 等多種智慧體，並有多家合作伙伴正在接入。

AI公司大力推廣可代使用者購物的智慧代理，但亞馬遜、摩根大通等平臺正以安全為由封堵這些代理，Target等網站的機器人驗證也讓代理時通時斷。The Verge記者實測Meta的Muse代理發現，任務常需反覆指導、結果不盡如人意，而在The RealReal上用代理批次鎖貨暴露出零和式搶購隱患。作者認為，購物中的發現樂趣與自主選擇權正被這種“最佳化”逐步消解。

本文基於Emerj播客對Poka、ABB和GE Vernova高管的訪談，探討工業AI與通用企業AI的根本差異，並提出四項關鍵洞察：為受控物理操作構建專用工業AI、連線運營資料以提供情境化指導、將AI助手嵌入工作流以加速一線問題解決，以及標準化工作流以實現多站點可重複部署。

Agentic CleanSlop 是 Product Hunt 上的一款工具，可用代理對你的應用進行身份驗證、掃描並修復問題。

輝達DSX AI工廠平臺將電網、設施與算力統一規劃與模擬，以應對電力成為基礎設施最大瓶頸的困境，GE Vernova、日立能源、施耐德電氣、西門子能源等企業已採用該平臺。

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：arXiv:2610.28901v1 Announce Type: new Abstract: We consider cooperative and iterative tasks involving multiple agents whose interactions are modeled as a dynamic game with shared constraints. We propose a general learning framework applicable to nonlinear and nonconvex dynamic games that uses previously executed feasible trajectories to construct a sampled joint safe set and player-specific cost-to-go functions. At each time step, the agents jointly select a previously sampled terminal state and a Generalized Nash equilibrium used to reach it. A new solution is adopted only if it satisfies a cooperative acceptance rule agreed upon by the players. The framework accommodates individual non-degradation requirements and selection criteria such as weighted costs and…

Weaviate 工程團隊在不到 17 小時內為全部客戶環境和基礎設施平臺完成補丁，以降低 eMudhra 證書頒發機構八張證書洩露帶來的風險，其中企業客戶在六小時內完成修補。

Uber宣佈在全公司範圍內部署ChatGPT Work，將這款智慧體平臺從最初1000人的試點擴充套件至全球逾3萬名員工，這是Uber首次面向知識型工作全員推出的智慧體部署。

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：Learn how to connect on-premises data sources to Amazon Quick privately with AWS PrivateLink, keeping analytics traffic off the public internet. This walkthrough covers VPC endpoints for user and API access, VPC connections for data sources, DNS and endpoint policies, identity federation, and enforcing VPC-endpoint-only access for compliance.

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：Many teams build a multi-agent AI prototype in days but hit a wall in production. This post defines five production challenges (orchestration, per-agent model selection, reliable operation, secure code execution, and monitoring) and shows how Deep Insight, an open source multi-agent system on AWS, solves each.

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：Vercel released durable-harness, an open-source package that logs every model and tool call so Pi, OpenCode, Claude Code and Codex sessions resume after crashes, timeouts and redeploys on Vercel, Postgres or a laptop. The post Vercel Open-Sources durable-harness for Crash-Proof Coding Agents appeared first on MarkTechPost.

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：The Adaptive Agentic Intelligence Framework (AAIF), built with Strands Agents, handles routine tasks locally with small language models and routes complex requests to cloud models through Amazon Bedrock. This post walks through building AAIF with local SLMs through Ollama and Amazon Bedrock, using an Internet of Things monitoring scenario.

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：Discussion | Link

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：Anthropic just announced an AI agent competitor to Meta's Muse and OpenAI's Dots: Claude Duet. The new agent, designed for professionals and small business owners, is launching as a "research preview," according to Anthropic, and it's available as a standalone app on iOS and Android and through Claude on the web. The AI agent is built around Claude Opus 5.5. Here's Anthropic's description of what it can do: Each morning, it sends you a short brief on the most important items across your connected apps. Throughout the day, it keeps tabs on those apps and messages you only when something needs your attention. You can also hand it tasks of a … Read the full story at The Verge.

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：Walmart will soon let AI agents shop on its website for you. Next month, the retail giant will allow you to connect your own AI agent to the site, with systems in place to let you approve and review purchases and set spending limits. The embrace of AI agents stands in stark contrast to Amazon, which blocked Perplexity's shopping agent and Meta's Muse AI assistant. But Walmart is opening its doors to AI shoppers - as long as they follow the retailer's rules. It has outlined a set of terms for agents, saying that "agents must not take actions beyond what the shopper intends" and cannot change a customer's payment method. Agents also can only … Read the full story at The Verge.

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：AT&T is working with OpenAI to make legal knowledge easier to access. Learn how its LegalEdge team is rethinking intake and scaling legal judgment.

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：Health & Bioscience

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：Train Muse Glimmer for free: Our recipe for out-betting Jev Join us for our inaugural conference, Forge 2026 Blog Train Muse Glimmer For Free Train Muse Glimmer for free: Our recipe for out-betting Jev PUBLISHED 10/20/2…

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：Training Muse Glimmer for free to out-bet Jev Join us for our inaugural conference, Forge 2026 Blog Training Muse Glimmer To Out Bet Jev Free Weights Club: Training Muse Glimmer to out-bet Jev PUBLISHED 10/20/2026 Table…

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：Featured Unstructured Unstructured Transform: Faster, More Accurate For Your Hardest Documents Oct 21, 2026 Featured Unstructured Unstructured Transform: Faster, More Accurate For Your Hardest Documents Oct 21, 2026 Aut…

Gerty 是 Product Hunt 上釋出的一款面向 Mac 的私有 ChatGPT 替代工具，主打在本地自動完成安裝配置，讓使用者無需繁瑣設定即可使用私密化的 AI 對話體驗。

HBO Max新劇《The Audacity》以黑色喜劇形式諷刺矽谷AI科技巨頭，由《繼承之戰》前編劇Jonathan Glatzer打造，主角是身家五億美元卻仍渴望“登上神山”的資料分析公司CEO。

Bloblings 是 Product Hunt 上的一個新專案，為每個 Claude Code 會話提供一個虛擬寵物。

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：Equating to the loss of 5% of its 93,000 global staff, job cuts come as bank announces jump in quarterly profits Business live – latest updates Barclays is cutting 5,000 jobs as part of wider cost-cutting plans that its bosses say are being driven by advances in AI. The job cuts will mean losing about 5% of the bank’s 93,000 global staff, with executives confirming they had already cut 1,000 workers since the start of the year. Continue reading...

Tetriz 工程智慧與聯結器在 Product Hunt 上釋出，主打將編碼會話與拉取請求、議題、質量指標和投資回報率關聯起來，幫助團隊衡量 AI 程式設計投入的實際產出。

加拿大出版社Les Éditions du Boréal已證實，暢銷海地裔加拿大作家Thélyson Orélien在創作小說《C'était ça ou mourir》時使用了人工智慧。Orélien此前多次否認相關指控，該作品也曾被撤出法國龔古爾文學獎候選名單。

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：Discussion | Link

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：How OpenAI works with distribution partners to uphold consistent safeguards, support responsible access, and clarify expectations in our Usage Policies.

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：Discussion | Link

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：Readers react to Simon Jenkins’ article that said it’s more important for children to learn how to speak fluently than study mathematics Simon Jenkins must have had a deeply unpleasant experience of learning maths at school because he seems to have a pathological antipathy towards it (We are teaching children maths they will never use. Isn’t it better to help them speak fluently?, 16 October). I have no qualms with his insistence on the importance of oracy. But his argument breaks down with the claim that it is either oracy or maths, but not both. This is nonsense. Oracy and maths skills should exist side by side. Both contribute to a person’s ability to realise their full potential, and to employment opportunities. Maths drives the understanding of the natural…

輝達GeForce NOW本週四迎來《使命召喚：現代戰爭4》首發，同步上線戰役、多人模式和DMZ撤離模式，並藉助RTX 5080級效能為終極會員帶來光追體驗。此外，Fire TV Stick使用者現可直接透過亞馬遜購買會員，史克威爾艾尼克斯的《最終幻想：共鳴》與《上古卷軸OL》新賽季等六款遊戲同步加入雲遊戲庫。

本文介紹如何使用 NVIDIA 開源工具 Nsight Python 在 Python 環境中剖析 GPU 核心。透過在程式碼中標註關注區域並用裝飾器包裝基準函式，開發者可以直接獲得按註釋區域、配置和執行次陣列織的 pandas DataFrame 硬體指標，無需再依賴 Nsight Compute 命令列報告。教程在免費的 Colab T4 上演示了安裝、單核心剖析、執行次數選擇、快取與時鐘控制、矩陣尺寸掃描、硬體計數器判斷計算瓶頸、核心融合測量以及 CUPTI 後端對比等完整流程。

Energize Lab 在沉默一個月後終於回應 The Verge 關於其 AI 機器人玩具 Eiliko 存在嚴重安全漏洞的報道，承認核心問題並承諾聘請獨立安全公司審查系統、刪除已儲存的對話記錄和使用者資料。然而，The Verge 與安全研究員驗證發現，該公司系統仍不安全：音訊和照片仍暴露在無保護的公共網址上，研究員仍擁有對機器人的完整 SSH 訪問許可權，可讀取對話記錄並修改玩具行為。

該論文提出 FLEX（靈活特徵編輯流水線），利用語言模型根據人類的自然語言糾正，迭代式地精煉機器人對任務特徵的內部表徵。與以往只允許“新增”特徵的編輯方法不同，FLEX 透過新增缺失特徵、泛化過於具體的特徵、遮蔽無關特徵三種互補操作，避免特徵集過度指定或範圍過窄。作者在二維導航、7 自由度 Franka 桌面操作任務以及真實機器人實驗中對 FLEX 進行了評估，結果顯示三種操作在互動場景中都不可或缺，且 FLEX 在保持表徵緊湊、可遷移的同時，取得了更高的歸一化獎勵。該工作已被 CoRL 2026 接收。

該論文提出Real2Sem2Real（R2Sem2R）流水線，僅憑一段單目RGB人類演示影片，就能為同一類別的多個語義物件例項生成數以千計的機器人演示資料。系統用DINOv3蒸餾的高斯潑濺跟蹤物體SE(3)位姿，並將由少量照片生成的“數字表親”與演示物體做語義對齊，再把手部動作對映為可行抓取，最後在IsaacSim或MuJoCo中批次生成資料。在兩臺真實機器人和四項任務上，用該方法訓練的策略在多樣物件例項上達到或超過遙運算元據訓練的策略。

Nuo Chen 與 Ya-Jun Pan 提出一個元學習框架，用於人機互動中的互動模式識別與自適應控制。該框架用元學習訓練的神經網路替代人工設定的切換閾值來判斷人類意圖，被動模式識別率由 90.1% 提升至 99.1%，主動模式由 79.8% 提升至 97.1%；同時引入自適應元學習控制器補償人類擾動，並用多軌跡元學習演算法聯合學習環境基表示與力相關引數，以應對未建模動力學。

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：arXiv:2610.28712v1 Announce Type: new Abstract: First-person humanoid teleoperation ties access to side and rear targets to operator rotation. We present coupled visual--motor remapping, pairing a panoramic azimuth warp with its inverse upper-limb yaw transform. A working prototype implements Global Zoom (GZ) and Focus+Context (F+C). In a within-subject study (N=8); 72 trials), GZ had the largest completion count (23/24; Fixed: 20/24; F+C: 21/24), whereas F+C was preferred by 6/8 participants. Therefore, completion and preference favored different mappings in this sample.

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯： Panache — RTS Belle Époque pour mobile Well this is delightful: I released Battle Drones yesterday under an Apache 2.0 License and there's already a French language fork which reskins the game to take place in steampunk 1900s France and adds an animated intro! Panache est un jeu de stratégie en temps réel pour mobile et navigateur, dans la France de la Belle Époque (vers 1900) : dirigeables, Fardiers à vapeur, Creusot cuirassés, automates en képi, Gargantua la forteresse marchante — et beaucoup de panache. You can play Panache here - it's worth clicking that link just to experience the new intro video. So far it's mostly just aesthetic updates as opposed to gameplay changes. The music has had a French twist added to it as well. The new steampunk models are so…

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：Meta says it's issuing a new class of shares to fund its AI initiatives. As part of its third-quarter earnings report released on Wednesday, the company says the non-voting stock will also go toward potential acquisitions and employee compensation, with trading beginning in 2027. "The scale of investment required to compete, sustain growth, expand infrastructure, and pursue new strategic initiatives amidst the AI transformation taking place in the industry is significant," Meta says in the press release. Susan Li, Meta's chief financial officer, added during an earnings call that the new class of shares will "expand our funding tools and gi … Read the full story at The Verge.