
Know AI. See What’s Next.
Slide 1 of 5: 姜宇辉的AI分身:知识助手与“另一个自己”之间
Latest news

姜宇辉的AI分身:知识助手与“另一个自己”之间 (opens in a new tab)
南方周末通过学者、用户与Second Me创始人的访谈及记者试用,观察AI分身的学术辅助、情感连接和商业化尝试,也呈现知识可靠性、人格边界、版权与隐私方面的疑虑。

智动未来据报完成亿元级Pre-A+融资,资金用于产线与服务机器人部署 (opens in a new tab)
机甲觉醒报道,智动未来宣布完成亿元级Pre-A+轮融资,由中美绿色基金领投、南曦创投联合领投,资金拟用于广州产线、服务场景部署和世界模型研发。“智子世界”仍属技术预览,商业化效果待更多应用数据检验。

范式智能介绍Phai-A1与Phai-M1,强调国产算力后训练优化 (opens in a new tab)
范式智能介绍面向代码与Agent场景的Phai-A1,以及面向真人影视的音画同步模型Phai-M1。公司展示了单卡运行和视频生成能力,并称模型在昇腾平台完成专项后训练优化。

腾讯Q3资本开支814亿元,称WorkBuddy日Token用量居国内首位 (opens in a new tab)
腾讯三季报显示营收2073.36亿元、资本开支814.33亿元。公司称WorkBuddy日活超1000万,按用户每日大语言模型Token消耗总量居国内首位;报道还讨论云业务增收与AI投入对盈利、现金流的影响。

黄仁勋离任后:硅谷101访谈英伟达的技术、供应链与组织挑战 (opens in a new tab)
硅谷101采访半导体、科技与投资人士,讨论黄仁勋离任后CUDA生态、系统整合、供应链、开源投入和组织传承的潜在变化。产业放缓、估值折扣及人才流失判断属于受访者观点。

Cola2探索连续隐空间语言建模,字母AI解读字节扩模路线 (opens in a new tab)
Cola2研究语义动态压缩、重构与生成平衡及连续—离散协同训练。文章据论文、媒体和匿名消息讨论字节Seed的技术与组织调整;“10万亿”项目、人事晋升及公司战略对应关系不能由论文直接确认。

蚂蚁与西湖心辰介绍Duo v1,探索全双工视听交互 (opens in a new tab)
Duo v1通过Intent与Realization模块衔接交互决策和音视频生成。文章报告30组交互、180项检查中通过168项,内容响应度为73.3%;复杂意图理解与长程一致性仍是团队后续优化方向。

Small-town merchants try AI assistants for menus and hotel analysis (opens in a new tab)
GeekPark interviews merchants in Rizhao, Chifeng and Chaoyang about Meituan's business assistant. Uses include menu cleanup, conversion analysis and room pricing. Platform figures and merchant accounts alone do not establish general AI-driven gains.

Mei Tao proposes agentic world models linking planning and feedback (opens in a new tab)
AI Tech Review outlines Mei Tao's AWM concept, connecting language models, multimodal representations and embodied feedback. Team demos illustrate the idea; visual consistency does not establish physical accuracy or reliable real-world execution.

Four Chinese AI models compared on investment research deliverables (opens in a new tab)
AI Tech Review tests Qwen4-Max, Hy4, Kimi and DeepSeek using frozen market data, comparing workpapers, Excel formula compatibility and API costs. Price forecasts remain unverified; this small test does not establish trading returns or predictive reliability.

Trying OpenSquilla’s model router: task tiers, model switching and costs (opens in a new tab)
The author tests OpenSquilla on project exploration and documentation. Reported cost and quality results depend on the benchmark and model pool; monthly tier recommendations still need validation for each use case.

Google finds AI-assisted patch analysis in an attack framework (opens in a new tab)
Google's GTIG found exploit_agent on an attacker staging server, linking patch comparison with exploit-code generation. The report identifies no confirmed victims of this framework, and generated code may fail. Defenders still need risk-based patching.

General AI models enter hardware development, from files to prototypes (opens in a new tab)
Youxin reviews hardware-design examples involving OpenAI, Anthropic and engineering tools, spanning circuits, enclosures, firmware and CAD. Some prototypes were built and tested, but digital checks, engineering review and physical reliability remain distinct hurdles.

Syncere raises $11M for Lume, a household robot shaped like a floor lamp (opens in a new tab)
Syncere uses a floor-lamp form for Lume: one robot handles some pick-and-place tasks, while two work together in Duet mode on tasks such as folding clothes. Demos are edited; pricing and delivery reflect the source date, and home reliability remains unproven.

Xiaohongshu tests Dada, turning personal expertise into interactive agents (opens in a new tab)
A media hands-on report uses Xiaohongshu's Dada beta to build a cooking assistant, exploring knowledge bases, tools, widgets, group chats and paid services. The product is still in testing; access may change and the commercial outlook is the author's interpretation.

EgoExplorer captures mobile manipulation with five trackers and a camera (opens in a new tab)
Shanghai AI Lab presents a wearable capture system using trackers, gloves and a camera, plus C³ motion reconstruction, and releases 100 hours of data and code. Reconstruction and multimodal generation are assessed separately; humanoid transfer is shown only in simulation.

ConvexLLM connects speech to language models through frozen text embeddings (opens in a new tab)
Researchers at CUHK-Shenzhen and Microsoft propose C-Gate, mapping audio into convex combinations of frozen text embeddings. The paper reports VoiceBench gains against its listed open-model baselines, while audio understanding and transcription still involve trade-offs.

An AI researcher's view of synthetic data, expert networks and outsourcing (opens in a new tab)
Jinyan Su draws on post-training research and job-search experience to discuss data vendors' tasks, sales and quality checks. This is personal observation and commentary; anonymous examples are not independently verified or representative of every provider.

Google unveils Gemini 4 Neon and expands Genie 4 access for eligible Ultra users (opens in a new tab)
Google introduces Neon for efficient, complex tasks, with a million-token context window and API pricing, and expands Project Genie interactions. Rollout is gradual; cybersecurity models and Genie access remain subject to eligibility and availability restrictions.

ChinaVenture: Xingchen Haoyu raises a Series B for outcome-based AI workers (opens in a new tab)
ChinaVenture interviews founder Zhang Lei about enterprise Agent Workers, results-as-a-service and overseas delivery. The reported funding is hundreds of millions of yuan. Revenue and customer claims come from the report and team; growth and robot products remain plans.