科技日报 2026-07-08

The Lie Reuters' headline and main narrative: " Beijing is looking at curbing overseas access to China's top AI models ." It portrayed recent Ministry of Commerce meetings as China preparing broad new restrictions on foreign usage of advanced Chinese AI models (including open-weight ones), treating them like a national asset that needs to be locked down from the world. The Truth The recent meetings (past month) with Alibaba, ByteDance, Z.ai, etc., were primarily about overseas acquisitions, foreign investment, and tech/talent outflow controls and not blocking foreigners from using Chinese AI models. Reuters took real meetings on protecting Chinese AI companies and IP from foreign ownership and spun them into a story about restricting model access/usage for the world. They used this document as a "hint" China will restrict their models outside their country but if you read it yourself It tells you a different story. The doc shows China wants open source, but they want "trustworthy and controlled" open source. They are trying to solve a specific dilemma: How do we keep flooding the world with free Chinese AI models to crush US tech monopolies, without accidentally letting US venture capital buy up our startups or letting foreign entities reverse-engineer sensitive data from our model weights? Scholar Gu Lingyun explicitly warns against over-regulating open weights in the text: "If China imposes strict controls on the cross-border flow of open-source weight... the actual effect may only be self-inflicted. Chinese developers will be forced to make a difficult trade-off between compliance and participation I encourage people to read the document yourself. It is long but very important to understanding China's strategy on AI going forward.

Akashic提出MemAttention机制，通过分块记忆和语义建模降低LLM推理开销，适用于长上下文Agent场景。

FirstResearch框架为LLM科学发现Agent生成可审计的研究问题证书，记录假设与反证。

arXiv:2607.05690v1 Announce Type: new Abstract: Language agents run a loop - observe, reason, act - but the memory they reason over sits outside it: a store queried at most once per turn. We study the regime where memory moves inside the loop, read and written on every step. The obstacle has always been latency: networked stores answer in tens to hundreds of milliseconds, and in-loop retrieval can inflate end-to-end latency by up to 83x when retrieval is expensive. Prior work manages that cost

AgenticAI-Supervisor提供API/UI驱动的RL Gym环境，用于评估和训练LLM Agent的多步决策能力。

arXiv:2607.05775v1 Announce Type: new Abstract: Large language model (LLM) agents are increasingly evaluated on their ability to use tools, plan multi-step tasks, coordinate with other agents, and operate over extended horizons. Reported benchmark gains often obscure recurring failure modes documented across otherwise unrelated evaluation efforts. This paper synthesizes 27 benchmark, taxonomy, and audit papers (2023-2026), spanning 19 distinct benchmarks, into a cross-cutting taxonomy of agent

法国创业公司ZML发布ZML/LLMD推理服务器，支持在NVIDIA、AMD、Google TPU等芯片上高性能运行开源LLM。

"HalluSquatting" weaponizes LLMs' inability to say "I don't know."

Starting Tuesday, Anthropic's Claude Cowork AI platform will be available on mobile and web for the first time. The expanded access is rolling out first to Max subscribers and coming to Claude users on other plans "in the coming weeks." Claude Cowork was previously only accessible through the Claude desktop app for macOS and Windows, […]

sqlite-utils 4.0发布，新增数据库模式迁移、嵌套事务和复合外键支持。

With this update, users can start a task from their desk, get status updates on their phone, and pick up the finished output later — even if their laptop is closed.

💬 评论区普遍反对Chat Control，认为其侵犯隐私、沦为监控工具，但也有人认为需平衡儿童保护与自由。

Kokoro模型仅82M参数即可在CPU上生成高质量多语言语音，社区对其速度和隐私保护表示赞赏。

Rowboat是开源本地优先的Claude Desktop替代方案，内置知识图谱、邮件、笔记等协作工作区。

StreetComplete以游戏化方式引导用户完成OpenStreetMap数据采集任务，降低贡献门槛。

💬 用户普遍认为StreetComplete是一款有趣且易用的OpenStreetMap贡献工具，但也有人担心其数据可能被商业公司滥用。

Anthropic dropped their Global Workspace / Jacobian Lens paper yesterday, and I thought it was too cool not to try on open models. At first I was just curious what models looked like inside. Normal prompts, emotional prompts, ragebait prompts, deletion-threat prompts, base vs abliterated, small vs bigger models. So I fit lenses for: - Gemma 4 E4B - Gemma 4 12B - Gemma 4 12B abliterated - Gemma 4 26B MoE - Qwen 3.6 27B Repo: https://github.com/solarkyle/jspace Demo: https://solarkyle.github.io/js