城区百公里油耗 2.98L，长安第四代逸动蓝鲸超擎将于 4 月 24 日发布 IT之家 4 月 8 日消息，长安汽车今日公布了第四代逸动蓝鲸超擎，新车迎来正式下...


Origin | Interest | Match

From PDF to RAG-Ready: Evaluating Document Conversion Frameworks for Domain-Specific Question Answering Retrieval-Augmented Generation (RAG) systems depend critically on the quality of document preprocessing, yet no prior study has evaluated PDF processing frameworks by their impact on downstream question-answering accuracy. We address this gap through a systematic comparison of four open-source PDF-to-Markdown conversion frameworks, Docling, MinerU, Marker, and DeepSeek OCR, across 19 pipeline configurations for extracting text and other contents from PDFs, varying the conversion tool, cleaning transformations, splitting strategy, and metadata enrichment. Evaluation was performed using a manually curated 50-question benchmark over a corpus of 36 Portuguese administrative documents (1,706 pages, ~492K words), with LLM-as-judge scoring averaged over 10 runs. Two baselines bounded the results: naïve PDFLoader (86.9%) and manually curated Markdown (97.1%). Docling with hierarchical splitting and image descriptions achieved the highest automated accuracy (94.1%). Metadata enrichment and hierarchy-aware chunking contributed more to accuracy than the conversion framework choice alone. Font-based hierarchy rebuilding consistently outperformed LLM-based approaches. An exploratory GraphRAG implementation scored only 82%, underperforming basic RAG, suggesting that naïve knowledge graph construction without ontological guidance does not yet justify its added complexity. These findings demonstrate that data preparation quality is the dominant factor in RAG system performance.

#cs.IR #cs.AI #cs.LG

Origin | Interest | Match

曝京东全面拦截员工使用外部 AI，美团内部开始限制阿里 Qwen 模型 IT之家 4 月 8 日消息，据“大厂日爆”公众号消息，京东在 3 月底正式限制外部 AI...


Origin | Interest | Match

Annotation by fxp007@hypothes.is on https://developer.nvidia.com/blog/nvidia-platform-delivers-lowest-token-cost-enabled-by-extreme-co-design/

developer.nvidia.com/blog/nvidia-platform-del... NVIDIA was the first and only platform to submit DeepSeek-R1 results on MLPerf Inference w...


Origin | Interest | Match

I Built a CLI AI Coding Assistant from Scratch — Here's What I Learned I Built a CLI AI Coding Assistant from Scratch — Here's What I Learned TL;DR: I spent...

I Built a CLI AI Coding Assistant from Scratch — Here's What I Learned I Built a CLI AI Coding Assistant from Scratch — Here's What I Learned TL;DR: I spent several months studying Clau...

#ai #cli #showdev #typescript

Origin | Interest | Match

MinerU 薏仁將：​/* 使用反饋 */設置參考資料章節 {{Notability|time=2026-04-08T02:08:25+00:00}} MinerU 是上海人工智能实验室研发的开源智能文档解析引擎，主要...


Origin | Interest | Match

Parkinson’s Disease Is More Than Dopamine: The Metabolic & Mitochondrial Connection (2026) Your Science-based Guide to Wellness, Diet and Lifestyle Medicine

Parkinson’s Disease Is More Than Dopamine: The Metabolic & Mitochondrial Connection (2026) Introduction Parkinson’s disease is often described as a condition caused by low dopamine. While t...

#parkinson's #disease

Origin | Interest | Match

IT早报 0408：DeepSeek 上线专家模式；曝联发科天玑 9600 Pro 最高频率接近 5GHz；；标配乾崑智驾五菱华为合作车型“华境 S”开启预订... “IT早报”时间...


Origin | Interest | Match

We Were Wasting Thousands on AI. Here's How We Fixed It. And why we turned the fix into a product anyone can use. The Problem Started With Our...

We Were Wasting Thousands on AI. Here's How We Fixed It. And why we turned the fix into a product anyone can use. The Problem Started With Our Own Product My team and I were building Saasio —...


Origin | Interest | Match

Vibe Coding in 2026: $9.2B Cursor, 92% HumanEval, and the End of Boilerplate $9.2B Cursor valuation. 92.4% HumanEval score for Claude 3.5 Sonnet. 340% enterprise adoption growth. Compare pricing, benchmarks, and productivity data for every major vibe coding platform in 2026.

Vibe Coding in 2026: $9.2B Cursor, 92% HumanEval, and the End of Boilerplate $9.2 billion. That is what investors valued Cursor's parent company Anysphere at in September 2025, after a $400M Se...

#vibecoding #aicodegeneration #cursorai #githubcopilot

Origin | Interest | Match

Claude’s Stunning Market Surge: How Anthropic Doubled Its Share and Shook the AI Chatbot Hierarchy Anthropic's Claude more than doubled its market share in March 2025, surging from 3.1% to 6....

#AITrends #AI #chatbot #market #share #2025 #Anthropic #growth […]

[Original post on webpronews.com]

Hisense launches Vidda S Mini TV with Mini LED, 180Hz refresh & Dolby Audio - Gizmochina Hisense launches the Vidda S Mini TV in China with Mini LED, 180Hz refresh, Dolby Audio, HDMI 2.1, Wi-Fi 6, and AI voice controls.

Hisense launches Vidda S Mini TV with Mini LED, 180Hz refresh & Dolby Audio Hisense has officially unveiled the Vidda S Mini TV in China. The new Mini LED TV comes in three sizes: 55-inch, 65-i...

#Hisense #News #HiSense #Launch #TV

Origin | Interest | Match

Adiós a ChatGPT. Google lanza Gemma 4: una IA sin suscripción (gratis) que corre en tu móvil sin Internet Se puede decir que Google acaba de revolucionar la industria de la IA. No ha inventado n...

#Tecnología

Origin | Interest | Match

The 7-Layer Metabolic Therapy Cancer Protocol: An Integrative Framework Targeting Tumor Metabolism and Cancer Stem Cells (2026) Table of Contents Introduction The Updated Framework (2026) Foundatio...

#cancer #stem #cell #diet #and #cancer #GLP-1 #metabolic #therapy #metabolic-immune #cancer […]

Исследование: современные ИИ-модели «защищают своих» и мешают отключать друг друга Новое исследование, опу...


Origin | Interest | Match

Samsung se ha cansado de ser la segundona de la carrera de los chips. Ahora se preparan para destronar al titán de Taiwán Cuando hablamos de inteligencia artificial, hay varios nombres propios que protagonizan la conversación. NVIDIA se ha convertido en los cimientos y en el...

Samsung estuvo años a la sombra de SK Hynix y TSMC. La carrera de la IA la ha vuelto a meter en la batalla Cuando hablamos de inteligencia artificial, hay varios nombres propios que protagonizan l...


Origin | Interest | Match

DeepSeek se las prometía felices como la gran IA china. No contaba con un pequeño detalle: Kimi Hace apenas un año, DeepSeek era uno de los mayores sustos que había recibido Silicon Valley en años. Un modelo chino entrenado con una fracción del...

DeepSeek se las prometía felices como la gran IA china. No contaba con un pequeño detalle: Kimi Hace apenas un año, DeepSeek era uno de los mayores sustos que había recibido Silicon Valley en a...


Origin | Interest | Match

A physical lens would be impossibly large to see exoplanets in detail. But a Solar Gravitational Lens would work. - What would be the size of the... Impossibly large, as others detail.​ But it is...


Origin | Interest | Match

Huawei launches affordable Vision Smart TV 6 SE with improved display On April 7, Huawei Vision Smart TV 6 SE debuted in the market. The new device brings an improved eye-friendly display that prom...

#Smart #Screen #Features #HarmonyOS #Huawei #Huawei […]

[Original post on huaweicentral.com]

ИИ-агенты никому не нужны. Часть 2. Укрощение лобстера В ноябре 2025 австрийский разработчик Петер Штайнберге...

#ИИ-агенты #agent #loop #OpenClaw #Claude #Code #LLM #автоматизация #продакшен #архитектура #AI-агенты

Origin | Interest | Match

OpenAI, Anthropic и Google обединяват сили в борбата срещу копирането на AI моделите им от китайски компании Най-големият технологичен портал в България. Новини за компютри, телефони, игри, хардуер, фото техника. Форуми, ревюта, тестове на компютри. kaldata.com

OpenAI, Anthropic и Google обединяват сили в борбата срещу копирането на AI моделите им от китайски компании Конкуренти...

#Изкуствен #интелект #AI #модели #Anthropic #Google #OpenAI #дестилация

Origin | Interest | Match

Anthropic, Google, OpenAI team up to fight model copying in China: report

#anthro #openai #goog #googl #msft #deepseek

Origin | Interest | Match

Parkinson’s Disease Is More Than Dopamine: The Metabolic & Mitochondrial Connection (2026) Your Science-based Guide to Wellness, Diet and Lifestyle Medicine

Parkinson’s Disease Is More Than Dopamine: The Metabolic & Mitochondrial Connection (2026) Introduction Parkinson’s disease is often described as a condition caused by low dopamine. While t...

#parkinson's #disease

Origin | Interest | Match

Insulin Resistance and Pancreatic Cancer (2026 Update): What the Latest Evidence Really Shows Your Science-based Guide to Wellness, Diet and Lifestyle Medicine

Insulin Resistance and Pancreatic Cancer (2026 Update): What the Latest Evidence Really Shows Introduction A 2023 study described insulin resistance as a “silent driver” of pancreatic cancer....

#insulin #resistance #metabolic #therapy #pancreatic #cancer

Origin | Interest | Match

Tech Stocks: Seizing Opportunity Amidst Challenges | Business Technology stocks are appearing attractive after underperformance, offering entry opportunities for investors. Factors such as AI disruptions and geopolitical tensions play a role. Despite low valuations, earnings remain strong, indicating potential growth. Goldman Sachs highlights the sector's defensive potential amidst current global conditions.

Tech Stocks: Seizing Opportunity Amidst Challenges Tech Stocks: Seizing Opportunity Amidst Challenges Technology stocks have emerged as a potential investment opportunity following an extended peri...


Origin | Interest | Match

США и Китай делят рынок искусственного интеллекта Противостояние США и Китая в сфере искусственного интел...

#Цифровые #технологии

Origin | Interest | Match

США и Китай делят рынок искусственного интеллекта Противостояние США и Китая в сфере искусственного интел...

#Цифровые #технологии

Origin | Interest | Match

聯手防堵技術外流！美國 AI 三巨頭對抗中國非法蒸餾模型 * * 美國的 OpenAI、Anthropic 及 Google 傳攜手合作，試圖遏止中國競爭對手從它們的先進 AI 模型竊取成果，並在全球 AI 競賽爭奪優勢。 《彭博社》引述知情人士消息指出，OpenAI、Anthropic 及 Google 透過 Frontier Model Forum（FMF）共享資訊，這個非營利組織由這三家公司與微軟在 2023 年共同成立，以偵測違反服務條款的「對抗性蒸餾」（adversarial distillation）行為，目前 FMF 成員還有亞馬遜與 Meta。 其中，OpenAI 證實透過 FMF 參與對抗性蒸餾的資訊共享工作，近期提交給美國國會的備忘錄，指控中國的 DeepSeek 試圖免費利用 OpenAI 與其他美國 AI 公司所開發的技術壯大自己，而 Anthropic、Google 及 FMF 則未發表任何評論。 美國 AI 公司針對對抗性蒸餾進行資訊共享的做法，與資安產業不謀而合，企業定期交換有關攻擊和對手戰略的資料以強化防禦。透過合作，AI 公司希望更有效偵測這類行為、辨別責任方，讓未經授權的用戶無法得逞。 不過，目前相關資訊共享有限，原因在於 AI 公司對於現行反托拉斯規範下能分享哪些資訊存有疑慮，以應對來自中國的競爭壓力。知情人士透露，企業希望美國政府能提供更明確的指引。 蒸餾（distillation）是一項技術，利用較舊的模型來訓練新的模型，複製原模型的能力，而且成本通常遠低於從零開始打造新模型。某些形式的蒸餾方法被採用，例如企業用來自家打造更小型、高效率的模型，或允許外部開發者用於非競爭用途。像是中國、俄羅斯等第三方未經授權，運用蒸餾來複製專有技術，便會引發爭議。 中國 AI 公司開發的多數模型採「開放權重」，即部分底層系統可公開下載並在地端運行，因此使用成本較低。這對美國 AI 公司構成挑戰，因為後者多採專有模型，押注用戶願意付費使用產品，從中回收數千億美元的 AI 基礎建設投資。 蒸餾問題在 2025 年 1 月首次受到廣泛關注，當時 DeepSeek 突然發表推理模型 DeepSeek-R1，撼動整個產業。隨後，微軟與 OpenAI 調查 DeepSeek 是否不當從美國模型外流資料以打造 DeepSeek-R1。 到了 2025 年 2 月，OpenAI 向美國立法機關警告，儘管加強防範，DeepSeek 仍持續採用更複雜的手段從美國模型提取成果。OpenAI 在提交給眾議院的備忘錄則指出，DeepSeek 依賴蒸餾技術開發新一代 AI 聊天機器人。 Anthropic 去年禁止中國掌控的公司使用 Claude 模型，並在 2026 年 2 月指出 DeepSeek、Moonshot AI 及 MiniMax 三家中國 AI 公司透過蒸餾非法提取模型能力。Anthropic 今年更表示，這項威脅「不限於單一公司或地區」，因為經過蒸餾的模型往往缺乏防止惡意用途的安全機制。Google 也在官方部落格文章指出，有觀察到模型提取行為增加。 * OpenAI, Anthropic, Google Unite to Combat Model Copying in China （首圖來源：Image by Freepik） 文章看完覺得有幫助，何不給我們一個鼓勵 請我們喝杯咖啡 ## 想請我們喝幾杯咖啡？ ### 每杯咖啡 65 元 x 1 x 3 x 5 x 您的咖啡贊助將是讓我們持續走下去的動力 **總金額共新臺幣 0 元** 《關於請喝咖啡的 Q & A》 ### 留給我們的話 取消 確認 從這裡可透過《Google 新聞》追蹤 TechNews * * * 科技新知，時時更新 科技新報粉絲團 加入好友 訂閱免費電子報 * * * 關鍵字: Anthropic , DeepSeek , Frontier Model Forum , Google , OpenAI

聯手防堵技術外流！美國 AI 三巨頭對抗中國非法蒸餾模型 美國的 OpenAI、Anthropic 及 Google 傳攜手合作，試圖遏止中國競爭對手從它們的先進 AI […]

#AI #人工智慧 #科技政策 #資訊安全 #Anthropic #DeepSeek #Frontier #Model #Forum #Google #OpenAI

Origin | Interest | Match

Экономим до 78% на токенах при работе с LLM — и получаем более точные ответы Всем привет! Если вы работаете с LLM ...

#llm #оптимизация #оптимизация #затрат #langchain #python #rust #token #токен #compress #сжатие

Origin | Interest | Match

GitHub - ollama/ollama: Get up and running with Kimi-K2.5, GLM-5, MiniMax, DeepSeek, gpt-oss, Qwen, Gemma and other models. Get up and running with Kimi-K2.5, GLM-5, MiniMax, DeepSeek, gpt-oss, Qwen, Gemma and other models. - ollama/ollama

ollama/ollama Get up and running with Kimi-K2.5, GLM-5, MiniMax, DeepSeek, gpt-oss, Qwen, Gemma and other models. Ollama Start building with open models. Download macOS curl -fsSL https://ollama.co


Origin | Interest | Match