Discover the AI Tools

抖音来客对酒店订单收取约12%的服务费，引发市场热议。官方回应称，目前没有付费推广机制，商家无法通过付费影响推荐结果，费用仅在交易完成后产生。这一政策旨在覆盖渠道服务成本，而非广告竞价。

OpenAI has completed a $7 billion stock buyback for employees, keeping its valuation at $85.2 billion. This move, which excludes external investors, suggests the company is not rushing into an IPO. Facing performance challenges and competition from Anthropic, OpenAI appears to be focusing on strengthening its position privately before going public.

Tech giants like Google and OpenAI promised AI would free us from work, even predicting a four-day week. But former employees reveal the opposite: AI has intensified workloads, with some logging 90+ hours during crunch times. Instead of more leisure, efficiency gains have become extra pressure to hit higher KPIs. Is AI truly liberating us, or just making us work harder?

Virgin Atlantic is rolling out ChatGPT Work across its teams, turning what used to take weeks of market research into just hours. By unifying data dashboards and acting as a digital chief of staff, the airline is speeding up decisions and improving the passenger experience. It's a clear sign that AI is becoming a standard tool in business operations, not just a tech experiment.

Alibaba's Qwen app has introduced paid plans for professional users, offering office assistant memberships from 19 to 128 yuan per month and separate pricing for AI video generation. The move signals a shift from free basic features to premium, high-frequency services, as the app continues to expand its AI capabilities.

Meta has released Muse Glimmer, a 30-billion-parameter open-source AI model designed to run on consumer hardware. It powers local AI agents for tasks like scheduling and email, keeping data private. This move aligns with Zuckerberg's vision of accessible superintelligence, though Meta still keeps its more powerful model closed.

NVIDIA 与 Apollo、贝莱德等六家金融巨头达成合作，计划筹集超 500 亿美元，为 AI 基础设施建设提供资金支持。这一举措不仅缓解了企业扩充算力的资金压力，也标志着 AI 竞争已从模型与芯片延伸至资本与基础设施层面。

Tencent has open-sourced BrowserSkill, a browser automation tool that lets AI agents work in a separate window without interrupting your browsing. It reuses your logged-in sessions, so no repeated logins, and works with popular agents like Claude Code and Codex. In tests, it handled tasks like summarizing Zhihu discussions and scraping social media comments in minutes. This could be the missing link for AI agents interacting with the real web.

Anthropic has inked a massive 20-year, $9.1 billion deal with Riot Platforms, a former Bitcoin miner, to secure 191 megawatts of computing power at its Texas facility. This move highlights a growing trend: traditional energy and crypto companies are repurposing their infrastructure to meet AI's insatiable demand for compute. As AI giants scramble for resources, this partnership could reshape how the industry powers its models.

DeepSeek has addressed the viral trend of its 'deep thinking' mode giving users nicknames like 'Momo' or 'the student who always asks why.' The company clarifies these are temporary context tags generated during reasoning, not stored personal data. The model uses them to tailor responses, but they vanish after the conversation ends. No hidden files, no long-term memory—just a clever way to keep chats on track.

TVB is teaming up with Capital Group to build a massive AI computing hub in Hong Kong, with the first phase targeting 10,000 PetaFLOPS by 2027. This move marks a traditional media company's bold entry into the AI infrastructure race, leveraging its industrial park and capital to offer advanced computing services. It's a sign that AI's backbone is attracting players from all corners.

Anthropic is quietly adding invisible watermarks to all Claude outputs, from text to images. This move, aligned with the EU AI Act, aims to make AI-generated content traceable. But what does it mean for you? Your emails, reports, and even casual chats with Claude now carry a hidden fingerprint. While it's a step toward transparency, it's not foolproof—heavy editing can erase the mark. Here's what you need to know.

NVIDIA researchers have found a way to let AI models share their 'memory'—the KV cache—so switching models no longer means starting from scratch. This breakthrough can speed up inference by up to 25 times, potentially saving costs and making model upgrades seamless. It's like passing notes between students, but for AI.

Alibaba's Qwen has launched an open platform for AI hardware, targeting developers and enterprises. The platform supports building apps for smart glasses, with features like natural language app creation and industry-specific tools. Early use cases include assisting the visually impaired and guiding cooking. The move aims to shift AI wearables from fixed functions to customizable, evolving intelligence.

Mark Zuckerberg's 6,500-word manifesto paints a rosy picture of AI for everyone, but critics argue it sidesteps the real harms already caused by the technology. With public trust in social media at an all-time low, the gap between his idealistic vision and the harsh realities of AI's impact on education, law, and society raises important questions about the tech industry's approach to AI accessibility.

Linux 7.2-rc7 is out, and it's unusually big. Linus Torvalds attributes the size to developers using AI tools to generate many small fixes. While he's not thrilled, he says it's acceptable. The kernel's strict review process remains key. Stable release expected next week, with distributions updating at their own pace.

抖音来客对通过豆包入口产生的酒店订单开始收取服务费，费率约12%。这一调整标志着AI入口与本地生活交易的深度融合，平台正探索AI流量转化后的商业模式。

优必选科技今日启动申购，成为A股市场首只人形机器人股票。发行价150.80元/股，市盈率高达219倍，引发市场热议。尽管中签率预计极低，但参考今年新股平均涨幅，单签收益或超20万元，投资者热情高涨。

8月10日，阿里云千问开放平台正式上线，面向手机、PC和AI眼镜三类终端全面开放服务接入。平台首批已覆盖物流、房产、本地生活、金融、汽车等十多个高频生活场景，用户无需在多个应用间切换，直接在对话窗口输入需求即可完成从咨询到下单的全流程。同时，平台为开发者提供标准化协议接入、一键授权、端到端调试等底层支持，并整合账号、支付、订单等核心环节。业内认为，这一举措不仅降低了用户多任务切换的摩擦，也标志着AI助手正从“信息问答”加速向“智能体”演进。

NVIDIA has unveiled its first open-source full-duplex speech dialogue model, VoiceChat 11B, which unifies speech recognition and generation in a single network. This breakthrough reduces latency to 448 milliseconds and supports simultaneous speaking and listening. It's a game-changer for real-time AI interactions, though it currently requires high-end hardware and is research-only.