deepseek对与搜索相关内容做了限制，报错：
Error: HTTP 400: Content Exists Risk (request_id: 80b1188d-3fa0-48e9-8b98-d09b17c6af1e)
我用的是hermes agent做的ai行业问题调研报告
到什么时候都有用心做事情的人
I am done with this shit. It is over. The state of engineering right now is horrible. It has been half a month since I started a new role at a big company. Nobody knows anything here. The specs, code, tests, PRDs, tickets, resolution of those tickets, reports, etc., everything is made by Claude Code. Nobody on my team likes this. They are being forced to ship as much as they can. I have heard multiple times from higher management that pushing code is not a bottleneck, so why are we slow? People are working 12 to 13 hours a day just to press enter. Nobody is reading anything. Humans in corporate are doing nothing on their own. Everyone, literally everyone, from an L1 to an L7 engineer here is doing the same thing. Talk to Claude. There is no sense of victory. Nobody is resolving bugs. In reality, nobody is thinking anymore. Everything is done by LLMs. It is so soul-sucking. I would not mind it, to be honest, if we were at least given the time to check out the code and see what is going where. But no, the goal is to just ship. No matter what happens.
muse spark在中国用不了，很遗憾
Muse Spark 1.3 is FREE on OpenCode right now 😳
you can use it with your favorite AI coding agent
getting started:
1. go to opencode.ai/zen
2. log in and get your API key
3. use base URL: opencode.ai/zen/v1
4. select Muse Spark 1.3 Free
5. start coding
> Model: Muse Spark 1.3 Free
you can also install OpenCode CLI:
npm i -g opencode-ai
free access is live right now 👀
正在构建企业制度本体，这个项目的方向非常匹配
A GitHub repo is built around one idea enterprise AI badly needs:
Your agent should not just know what is true.
It should know what was true, when it was true, when the system learned it, and what changed afterward.
Utopia calls itself an open-source enterprise world model. In practice, the interesting part is a bitemporal knowledge graph at the center of the system.
Most RAG systems are optimized for
→ find relevant documents
→ retrieve passages
→ answer the question
But enterprise knowledge is constantly changing.
> A customer changes status.
> A contract gets amended.
> An employee changes teams.
> A policy gets replaced.
Yesterday's fact becomes today's outdated fact.
Utopia does not simply overwrite the old knowledge.
Every extracted fact can carry two timelines
→ when it was true in the real world
→ when the system came to know it
If a fact is corrected, the previous version is closed and linked to the new one instead of disappearing.
So an agent can ask things like
> What did we know about this account three months ago?
> What changed between January and March?
> Where did this fact come from?
> Was this asserted directly, or derived from other facts?
Utopia has an editable ontology, entity resolution, conflict detection and rule-based derivation. Derived facts remain marked as derived and retain the facts they came from. When new information conflicts with old information, the system has explicit ways to resolve it rather than silently replacing whatever was there before.
It still has the familiar AI stack underneath
→ PDFs, docs, spreadsheets, Markdown, HTML
→ GitHub, Jira, RSS and S3 ingestion
→ Tantivy full-text search
→ pgvector
→ RRF retrieval fusion
→ OpenAI-compatible models
→ MCP tools
→ Postgres + a Rust backend
But the graph changes the role of retrieval.
Normal RAG asks -
What information is relevant right now?
Utopia is trying to answer something harder -
What did the organization believe, when did it believe it, why did it believe it, and how did that knowledge evolve?
That is a much more interesting foundation for enterprise agents than endlessly dumping today's documents into a vector database.
Here's the GitHub Repo: github.com/deeplethe/utopia
这个节奏有一些吓人，从没有哪家公司频发的发布大模型，希望不是为了刷存在感
Google 今天发布了 Gemini 3.8 Flash，同时推出的还有一个专门用于网络安全的变体 3.8 Flash Cyber。
Google 最近这个发布节奏很有意思。7 月底发 3.6 Flash，8 月中发 3.7 Flash，今天又发 3.8 Flash，间隔都只有大约三周。而与此同时，今年 6 月本该发布的 Gemini 3.5 Pro 至今没有出现。Google 在 Flash 这条线上疯狂迭代，但更高端的 Pro 系列迟迟没有动静。
再加上 8 月初 DeepMind 经历了一轮震荡，创始人兼 CEO Demis Hassabis 从日常管理退出，转任 DeepMind 主席和 Alphabet 首席科学家，首席科学家 Jeff Dean 也离职创业。新任 SVP Koray Kavukcuoglu 接管了 Gemini 的开发。这种背景下连续推 Flash，不知道是不是某种程度上的刷存在感。
回到模型本身。
3.8 Flash 和 3.7 Flash 一样的价格，每百万输入 Token 0.75 美元，每百万输出 Token 3.75 美元。
注意这个价格是促销价，2026 年底到期后翻倍变成 1.5 和 7.5 美元。
跑分上，3.8 Flash 在 DeepSWE v1.1（一个衡量模型自主解决复杂工程问题能力的评测）上拿到 73.7%，仅比 Claude Opus 5 的 74% 低 0.3 个百分点，超过了 GPT-5.6 Sol 的 72.7%、Claude Sonnet 5 的 53.8%，以及上代 3.7 Flash 的 65.3%。
至于为什么评分高，Google 自己的解释是 3.8 Flash “works harder”，遇到复杂任务会多想几步、多调用几次工具，代价是消耗更多 Token。
另一个跑分细节：3.8 Flash 在 Gray Swan 提示词注入（prompt injection）评测中被攻破的概率只有 5.5%，对比 DeepSeek V4 Pro 的 60.1%、Grok 4.6 的 51.8%。也就是说，用 3.8 Flash 做 AI Agent 被恶意指令劫持的风险要低得多。
跟着 Gemini 3.8 Flash 一起发布的还有 3.8 Flash Cyber，这是一个专门为网络安全防御训练的模型，只通过 Google 新推出的 Fairwind Program 向经审核的政府机构、关键基础设施运营商和安全研究人员开放，目前全球有 650 多个合作伙伴。普通开发者用不了。
Google 把这个模型的能力框定在安全防御上，强调它擅长找漏洞和修漏洞，而非攻击利用。在 CyberGym 漏洞发现评测中，3.8 Flash Cyber 拿到 86.2%；在覆盖 20 种编程语言的内部评测中，漏洞发现成功率超过 70%。在 CWE-Bench 自动修补评测中，它以 47.2% 的 pass@1 逼近了前沿模型的 47.8%，但成本低得多。
一些实际应用案例：
Chrome 安全团队发现 3.8 Flash Cyber 比最好的商用大模型多修复了 2.6 倍的 Chrome 漏洞；
安全公司 Wiz 的内部渗透测试中，它的召回率高出 7.5% 到 9.7%，成本却低了 2.3 到 5.2 倍；
Google Cloud 漏洞研究团队用它在不到两小时内发现了一个关键漏洞，这类漏洞通常需要数月的人工研究。
3.8 Flash 已经可以通过 Gemini API、Google AI Studio、Android Studio、Antigravity 等渠道使用，Google AI Pro 和 Ultra 订阅用户在 Gemini App 中也能用上。企业版通过 Gemini Enterprise 接入。Cyber 版需要申请 Fairwind Program。
这个创意好哈，魔法哈
Reddit 社区的一位用户开发了一款
能让 AI（Claude）在手写板上与用户“笔对笔”
进行互动学习的软件
有点像《哈利·波特》里汤姆·里德尔的日记一样“能手写回信”和书本交流...
你在屏幕上用手写笔写下想法或笔记，Claude 就会直接在同一页面上用手写字体给你“写回信”，进行实时互动和探讨。
核心功能：
双向手写互动： 你在空白页面上手写记录、发散思路，Claude 识别后在合适的位置直接写下反馈、追问或拓展建议。
文献与电子书深度研读： 支持导入 PDF 和电子书。你在书页上划重点或写批注，Claude 会针对标记内容主动生成旁注，像私人教授一样与你讨论，甚至还能根据材料主动给你出题测验。
沉浸式复利学习： 摆脱了传统打字聊天的机械感，让学习过程更慢、更具探索性，强化记忆与理解。
完了，被美国人占领了哈
OX Alpha is hosted in America!! 🇺🇸
I sent 49 identical one-token requests from 7 Cloudflare regions, rotated the order to control for load and matched every response to OpenRouter’s per-generation server timing.
This isolated the geographic network floor:
ATL: 57 ms
DFW: 76 ms
Europe: 125–132 ms
Asia: 202–340 ms
Australia: 279 ms
A robust 6-of-7 fit converged at 43°N, 73°W, with the plausible band covering the US East Coast into southeastern Canada.
Best interpretation: OX Alpha’s serving path is anchored in eastern North America, most likely US East.
The only potential remaining caveat: this could be @OpenRouter backend handoff rather than the physical GPU rack.
semantica的含金量被严重低估了
semantica
Graph-Native Infrastructure for Context and Accountable AI Systems.
github.com/semantica-agi/sem…
这套餐太值了
卧槽，原来现在还是能以每个月 99 美元的价格，订阅 Grok Harvey 这个级别的会员，这个超值！
你相当于一下获得了：
1. 价值 300 美元的 Grok Harvey（包括 Grok Build 权限）
2. 价值 200 美元的 Cursor Ultra 会员，还有 Fable 5 的额度。
3. Grok Bot 的独立额度（这三个额度都是独立的）
4. Grok Image 和 Video（图像与视频模型）的额度
5. 会自动送你一份 Twitter Premium+ 订阅
折合下来只要 700 块钱一个月，简直夯爆了！
尤其是 Grok 后面还有更强的模型要发布。
可以看一下具体的使用路径：
1. 准备 Grok 账号：先登录一个 Grok 账号。
前提条件：如果你之前是用 Twitter（X）登录并订阅的，需要先取消 Twitter 订阅（取消后不会立即生效，所以我建议直接注册申请一个新的 Grok 账号。虽然新账号没办法共享你原有的 Twitter Premium，但这次订阅）。
如果是用旧账户或者 Twitter 授权登录，操作起来会有点麻烦，所以强烈建议用新账户注册。
2. 订阅促销活动：
先去第一个链接(grok.com/supergrok/targeted-…)，订阅获取两个月的 Super Grok 促销。
订阅成功后，再去下面第二个地址(grok.com/supergrok/targeted-…)，订阅连续三个月 99 美元的 Super Grok Harvey。
注意：后续记得及时取消订阅，否则到期后会恢复成 300 美元/月。
3. 绑定 Cursor 会员：
下载并登录 Grok Bot，登录时有个选项会要求选择 Cursor 授权登录。
这个时候，只要你有一个相同邮箱的 Cursor 账号（如果没有，就用这个邮箱注册一个），系统就会自动给你充值 Cursor 的 Ultra 会员。
