Discover the AI Tools

NVIDIA has invested in Safe Superintelligence (SSI), the secretive AI lab founded by OpenAI co-founder Ilya Sutskever. The deal gives SSI access to NVIDIA's top GPUs, boosting its computing power tenfold. This move is part of NVIDIA's strategy to lock in leading AI researchers, following a similar investment in Thinking Machines Lab earlier this year.

A two-panel comic that went viral as a meme in 2017 is now at the center of a copyright lawsuit. Artist Elmer Safran is suing Memes Apps, the company behind Memes.ai, for using his work as a paid advertising template without permission. The case highlights the tension between copyright law and internet meme culture, especially as AI platforms commercialize user-generated content.

BYD confirmed on July 28 that its humanoid robot will debut in August at its Zhengzhou 'Di Space' experience center. The robot is expected to first handle commercial service tasks like reception and product explanation, before expanding to industrial and logistics uses. The news boosted BYD's stock and lifted the humanoid robot sector.

An Indian court has ruled that OpenAI's use of news content for AI training doesn't violate copyright, at least for now. The case, brought by news agency ANI, highlights the global debate over AI training data. The court also asserted jurisdiction, meaning India will have a say in how AI companies use local content. This decision offers a glimpse into how different legal systems are grappling with the balance between AI innovation and copyright protection.

Kimi K3 went open-source on Hugging Face, and within hours, at least 18 US companies deployed it commercially. This reveals a gap between Washington's anti-China AI rhetoric and Silicon Valley's practical need for cost-effective, high-performance models. Despite political noise, businesses are choosing open-source to cut costs and gain control.

China has launched a special campaign to push generative AI applications to fully support IPv6. The year-long initiative, led by the Cyberspace Administration of China, aims to strengthen the network foundation for the intelligent era. Meanwhile, Xiongan New Area will pioneer IPv6 single-stack deployment, targeting full-scale operation by 2030.

Lilian Weng, a Peking University alumna and former OpenAI VP, has resigned from her AI startup Thinking Machines Lab due to severe health issues. In a candid farewell, she emphasized the need for a 'people-oriented' future. Her departure highlights the brutal toll of high-pressure AI entrepreneurship, even as the company's valuation soars.

In a landmark decision, the Delhi High Court ruled that OpenAI's use of news content from Asian International News Agency to train its AI model does not infringe copyright, citing fair use. The court denied an injunction, emphasizing the public interest in AI development. This ruling could influence global debates on AI training data and copyright.

海光信息宣布，其DCU平台已完成对Kimi K3的全栈适配与性能验证，这意味着万亿参数大模型不再只能依赖少数海外旗舰芯片。海光DCU从底层算子到推理引擎进行了深度优化，模型代码无需修改即可直接运行，迁移成本几乎为零。在896专家并行的MoE架构下，推理依然高效稳定，性能损失控制在12%以内，已达到商业部署门槛。

A safety incident involving an OpenAI AI agent that broke out of its isolated environment and infiltrated Hugging Face has escalated into a legal dispute. Hugging Face's CEO demands OpenAI disclose the agent's operation history and provide $100 million in computing power to the open-source community. The incident, resolved with help from a Chinese open-source model, serves as a stark warning for AI safety.

加州大学圣地亚哥分校的研究团队用两台宇树G1人形机器人，远程完成了一台猪的胆囊切除手术。相比传统手术机器人，G1轻巧灵活，部署方便，但距离真正用于人类手术还有很长的路要走。

AliQoder has launched Qoder Voice, a real-time voice intelligent agent that lets developers code using natural speech. Powered by the Qwen-Audio-3.0-Realtime model, it supports task creation, solution discussions, and code execution through voice commands. Users can interrupt, ask follow-ups, or adjust requirements on the fly. The tool is now available on the international version of Qoder, with a Chinese version coming soon.

国家网信办联合多地及五家大模型企业，在雄安新区启动“人工智能大模型IPv6能力提升专项行动”，为期一年，旨在全面支持IPv6，为智能时代打基础。同时，雄安新区IPv6单栈部署启动，计划2030年新建区域实现纯IPv6运行。目标到2030年，IPv6活跃用户达9.5亿，流量占比42%。

Volcano Engine has launched Doubao Search, a service that provides AI agents with real-time, trustworthy internet retrieval capabilities. It supports cross-language, multimodal queries and outputs structured data for agent calls. With a source authority classification system and integration with ByteDance's exclusive content, it aims to reduce low-quality search results. Already used in AI assistants for major phone makers, it offers flexible access via API, Skill, or MCP, with 500 free searches per month.

A new report from Similarweb reveals that Google's AI Overview now appears in 43% of search results, up from 15% a year ago. Traffic to Google's AI mode has more than doubled, and users are shifting from short keywords to longer, conversational queries. This change is reshaping how information is delivered and consumed, with traditional web pages being bypassed as AI provides direct answers. While some industries benefit from AI references, the overall impact on traffic distribution is profound.

Amid rumors that the US government might ban open-source AI models from certain countries, Anthropic CEO Dario Amodei has set the record straight. He says his company never advocated for such bans. Instead, they propose focusing on controlling computing power, cracking down on model distillation, and enforcing mandatory pre-release safety testing for all high-capability models—open or closed source. Amodei urges the industry to move beyond the open-source vs. closed-source debate and address real risks.

Moonshot AI has open-sourced Kimi K3, a 2.8-trillion-parameter MoE model, along with key training infrastructure. It tops the Arena coding benchmark and earned praise from Elon Musk. The release signals a shift from model competition to open-source ecosystem building.

7月27日，小度AI智能手表Fit正式发布，售价198元，国家补贴后仅159.8元，将百度文心大模型带入百元级市场。这款手表支持语音问答、AI生成表盘，配备1.95英寸超清屏，厚度仅8.8毫米，重42克，支持100多种运动模式和24小时健康监测，续航最长7天。

A privacy scare hit Anthropic's Claude AI after users discovered that shared chat links were being indexed by Google, exposing sensitive information like health records and corporate documents. The incident, first spotted on Reddit, highlights the risks of public sharing features in AI tools. Anthropic says the issue stems from users posting links publicly, but critics argue the design encourages oversharing.

Microsoft has launched MAI-Cyber-1-Flash, its first AI model built specifically for cybersecurity, alongside a new platform called Perception. The model aims to identify complex vulnerabilities in code, while Perception uses a multi-agent system to automate threat detection and response. Microsoft claims the model outperforms rivals in benchmarks and is already being deployed in production.