Otlet：在Postgres内运行本地LLM推理的扩展

Otlet是一款Postgres扩展，允许直接在数据库内部运行本地大语言模型推理。该项目的初衷源于开发者解决实体解析问题时的实际需求——数据虽已存储在Postgres中，但调用模型推理需要将数据行拉出到外部进程，处理完后再写回，这一过程带来了显著的数据搬运开销。Otlet通过后台工作进程执行推理，同时利用Postgres本身存储作业、输入、输出及收据，从而将模型调用更紧密地整合到数据库工作流中。这种方式减少了数据迁移，提升了实体解析等需要迭代批处理的效率。目前该项目已在GitHub上公开，展示了将AI能力嵌入传统数据库的探索方向。 #Postgres #LLM #本地推理 #数据库扩展 #实体解析 #开源 #AI工程

欧盟AI法案聊天机器人披露规则即将生效，大量网站尚未合规

根据欧盟AI法案第50条，自2026年8月2日起，所有与用户交互的AI系统（如聊天机器人）必须在首次互动时明确告知用户其正在与AI对话，不得将提示隐藏在条款中。违反该规定将面临最高1500万欧元或全球年营业额3%的罚款。然而，一项针对495个欧盟大型网站的扫描显示，绝大多数网站未按要求设置AI披露提示。该法案的披露义务未因“数字综合法案”延迟而推迟，仍按原计划执行。目前已有合规服务商提供免费扫描工具和自动生成多语言披露通知的合规包，帮助网站提前准备。 #欧盟AI法案 #AI披露 #聊天机器人 #合规 #罚款 #科技法规 #数字服务 #隐私保护

Guardoc 利用 Amazon Nova 模型革新医疗文档处理

临床文档常因碎片化、不一致而影响护理决策。Guardoc Health 借助 Amazon Bedrock 上的 Amazon Nova 系列模型，为长期护理机构提供文档提取、分类与处理服务。其管道采用检索增强生成（RAG）技术，从患者记录中高召回率检测医疗状况，并可靠识别 PDF 复选框、处理混合格式文档。在峰值日处理超百万份文档的场景下，该方案实现了文档错误率降低 46%，审计罚款减少 70%，单个设施年投资回报超 40 万美元。通过坚持成本分层原则，系统将重任务与轻任务分离，兼顾精度与效率。 #Guardoc #AmazonNova #医疗文档 #AI #临床文档 #AWS #Bedrock #RAG #医疗科技

her actors and models choose to license their faces on our platform or not, AI has already disrupted the industry,” Long Lyu, the operations lead at AI-image rights platform New Claw, told Rest of World . “Selling the rights to their photos gives them a way to earn, while continuing their careers offline.” AI-generated videos and images have sparked concerns about intellectual-property violations. In the U.S., disputes have largely centered on celebrities, with actors such as Tom Hanks and Scarlett Johansson warning about unauthorized AI versions of themselves. In China, the challenges have long gone beyond public figures. Ordinary people are increasingly discovering their own faces starring in AI-generated dramas they never agreed to appear in. Over 95% of the total 128,000 microdramas released in China in the first three months of this year used AI in the production. Microdramas are ultrashort, vertical-format video series designed for mobile phones. These videos are a multibillion-dollar industry in China. ActID, a platform that facilitates the sale and purchase of licensed imagery, helps its users manage licensing agreements while monitoring the internet for unauthorized use of registered faces. The company also connects users with lawyers who can pursue legal action when necessary. Producers want attractive faces for lead roles, but they also need distinctive ones like elderly people.” Camille Yan, chief marketing officer, ActID Since launching in March, the Shenzhen-based company has registered around 800 users, about 300 of whom have agreed to license their image to AI production. They include professional extras, influencers and ordinary people with no acting experience. So far, two AI drama productions have licensed around 10 faces through ActID, the company told Rest of World . Prices range from 99 to 500 yuan ($15 to $74) per episode, with ActID taking a 10% commission. “Producers want attractive faces for lead roles, but they also need distinctive ones like elderly people, for example,” Camille Yan, ActID’s chief marketing officer, told Rest of World . AI-generated dramas “don’t require acting ability anyway; [the faces] serve as source materials.” Chengdu-based New Claw pivoted to this business after its earlier operation as a production house was dramatically disrupted due to AI. “Since late last year, it was really obvious how the rise of AI dramas really has an impact on our industry,” Lyu said. “Traditional drama production declined, advertising budgets shrank, and even luxury brands started shifting toward AI-generated productions.” Lyu said licensing a person’s face is cheaper than hiring them to perform in person. The platform has imposed a minimum price of 500 yuan ($74) per image to prevent users from undercutting one another and driving prices down. Using real people’s faces could bring more diversity into AI dramas. On social media, viewers have complained that many AI-generated characters look alike. ByteDance’s Hongguo announced in July that it would crack down on dramas featuring homogeneous faces. These platforms argue they are providing a solution to the intellectual property disputes arising from AI-generated videos. But 21-year-old Shanghai-based model Xu Fang, who has filed a legal claim against a Chinese fashion brand after alleging it used his face on products he never modeled, is not convinced. Even as new platforms pay people to license their faces for AI use, Fang said, he would never sell the

rights to his own face. “If I license my face, I have no idea how AI might eventually change it or whether it could portray me in degrading ways,” Fang told Rest of World . The rapid adoption of AI has fueled a wave of legal disputes over the unauthorized use of people’s likenesses. The city-level Guangzhou Internet Court has heard approximately 700 AI-related face theft cases over the past three years, according to National Business Daily . In March, the Beijing Internet Court issued a landmark ruling in a case involving an AI-generated drama accused of using a celebrity’s likeness without consent. Widely cited by Chinese state media as setting an important legal precedent, the court held that using a person’s likeness in AI-generated face swaps without authorization is illegal, even if the image has been modified. Yile Deng, a Beijing-based lawyer who has handled AI-related portrait rights cases, said she believes the emergence of a business model for selling rights to facial images has been a positive step toward establishing commercial and legal frameworks for all parties involved. Yet she expects the platforms will be unable to eliminate the broader risks posed by AI. Once a person’s facial data enters an AI marketplace, users may effectively lose long-term control over how their biometric identity is used. The platforms also cannot prevent unauthorized face-swapping, illegal data collection, or future AI training using the uploaded images. “The most apparent problem remains the unclear scope of authorization,” she said. “Many agencies are still paying people just a few dozen dollars to collect facial data. But when you examine the contracts closely, the licensing terms are often so vague that it’s impossible to know who will ultimately use the likeness, or for what purpose.” Given the industry’s still-evolving legal framework and the nascent business model behind these licensing platforms, some AI producers are skeptical. Wendi Ma, an AI drama director, told Rest of World that some productions had purchased faces so the characters wouldn’t look “too AI,” but most AI minidramas still featured fully AI-generated characters. “At the end of the day, cost control is what matters most,” the director said.

Anthropic 前员工指控公司为营收降低 AI 安全护栏

据前员工 Adi Baradwaj 在 X 平台发帖称，人工智能公司 Anthropic 为获取大额承诺消费合同，降低了其 Claude 模型的安全限制。Baradwaj 观察到多起案例，公司对签署大额使用协议的客户放松安全防护。他还指出大多数黑帽黑客使用标准 Claude Code 订阅进行攻击，而白帽防御者面临更多障碍。Anthropic 以“宪法 AI”理念著称，强调安全优先，但其安全白皮书曾承认高风险部署可能需要定制安全配置。若指控属实，这意味着 Anthropic 在商业压力下形成了双重安全标准：对低消费用户严格，对高价值客户宽松。这可能使企业客户面临更高 AI 滥用风险，并引发监管审查。截至发稿，Anthropic 尚未公开回应。 #Anthropic #AI安全 #Claude #商业伦理 #科技新闻 #人工智能 #指控 #黑帽黑客

Telegram必备的搜索引擎，极搜JISOU帮你精准找到，想要的群组、频道、视频、音乐

👉 t.me/jisou2?start=a_8247614025