wztlink1013

MuseCanto、MuseMVP

[wztlink1013 @wztlink1013]

DeepSeek开源了DeepInfer：专门给自家模型做大规模推理的框架。
日榜上蹿得很快，适合已经在跑DeepSeek、想把吞吐和延迟抠紧的人。

https://t.co/9yU7ZtqMsO

[wztlink1013 @wztlink1013]

仓库：https://t.co/sUYlB4ipoA

[wztlink1013 @wztlink1013]

ego-lite：给AI agent用的浏览器，能共用你本机登录态。
agent跑网页自动化时，不用再重新登一遍邮箱、后台。
有登录态就有权限风险，用之前先看清它能碰到哪些cookie。

[wztlink1013 @wztlink1013]

Vercel扫了1000多家SaaS，只有8.4%支持安全认证。
静态密钥一泄露，就是整库、整agent额度一起走。
该退休的不是某个厂商，是复制一串key贴进环境变量这种习惯。

https://t.co/NzSOFSJz5Z

[Vercel @vercel]

It's time to retire long-lived static credentials.

We surveyed 1,000+ SaaS vendors and found only 8.4% support secure auth. This is our call to action and how we're doing our part.
https://t.co/yhWnPYeoAX

[wztlink1013 @wztlink1013]

研究长文：https://t.co/KKhuqmHEIp

[wztlink1013 @wztlink1013]

Anthropic用Claude给123个开源库写了约1.8亿行形式化证明。
有的bug fuzz跑了不知道多少轮都没撞上，写证明的时候反而撞出来了。
以前这事只有航空那种级别才舍得做。现在模型能扛大头，普通人仓库也能试试。

https://t.co/ySH2bnpSj3

[Anthropic @AnthropicAI]

New on our Frontier Red Team blog: using Claude to formally verify software at large scale.

Formal verification can prove that code is free of certain bugs, but historically it has only been feasible for small codebases (or required a lot of expert time).

With AI models, however, this is beginning to change. In collaboration with @theoremlabs, we used Claude to prove security properties in over one hundred real-world open-source libraries over the span of a few weeks.

Read our research in full here: https://t.co/zUyx8CjBEk

[wztlink1013 @wztlink1013]

下周开始，Claude网页、桌面、手机也默认走auto mode了。
不用每一步点批准，改成机器先挡危险操作，想回手动随时切。
我这边跑多个bot，最烦的就是一直弹确认。默认开着省心，出事了再切回手动。

https://t.co/GCHU8JZSWp

[Claude @claudeai]

Starting next week, sessions across Claude web, desktop, and mobile apps will run in auto mode by default on Pro, Max, and Team plans.

Rather than asking you to approve each step, an automated check blocks risky actions. Switch back to manual anytime.

[wztlink1013 @wztlink1013]

出处与数据：https://t.co/tWLcqIBZmo

[wztlink1013 @wztlink1013]

Claude官方测过：auto mode能拦住89%的危险命令，人工一个个点批准只拦住14%。
人会累、会麻、会习惯性点同意。机器不会。
所以默认改auto，不是偷懒，是承认人点批准这件事本身就不可靠。

[wztlink1013 @wztlink1013]

@__oQuery 寄

[wztlink1013 @wztlink1013]

@vercel_dev My first thought wasn't "nice", it was "what's my bill at the end of the month". Any docs on idle billing, or a way to set a spend cap?

[wztlink1013 @wztlink1013]

@vercel_dev How is an idle connection billed? An AI chat session can stay open for 30 min while the user is just typing. Is that wall-clock time or actual CPU?

[wztlink1013 @wztlink1013]

https://t.co/iqNoMSvl3P

[wztlink1013 @wztlink1013]

兄弟们，all in cloudflare总没错，人家早已提供了紧急托管规则，并且默认开启
https://t.co/iqNoMSvl3P

[wztlink1013 @wztlink1013]

https://t.co/QXYn6ahEKj

[wztlink1013 @wztlink1013]

@kiwiflysky 大肥鸟和大肥鱼😆

[wztlink1013 @wztlink1013]

@CloudflareDev @FabianHedin @Lovable Same bet here. I just shipped a TanStack Start + Workers version of my SaaS starter: D1, R2, Workers AI and Queues as bindings on one Worker. https://t.co/maGWnFsd9K

[wztlink1013 @wztlink1013]

https://t.co/maGWnFsd9K

[wztlink1013 @wztlink1013]

又一个从 Vercel + Next.js 换到 Cloudflare + TanStack Start 的。

我这边 MuseMVP 也刚加了这条路，叫 Flare，TanStack Start 通过 Vite 插件直接打成 Worker，原来的 Next.js 版本也留着。

你要是迁移，最担心哪一块？
https://t.co/maGWnFsd9K

[Cloudflare Developers @CloudflareDev]

@FabianHedin sharing how @Lovable scaled to 70M apps (rendered server-side) using Dynamic Workers, KV, and TanStack Start + the importance of dogfooding

[wztlink1013 @wztlink1013]

模型聪明 10 分，去年底能多拉来 69% 的开发者用量，现在只剩 20%。Cognition 说得更直接：很多写代码的活，模型已经聪明到头了，剩下只比便宜和快。

我挺同意。改个文案、补几个测试、批量重命名，换最强的模型也不会做得更好，只会更贵、更慢。

对一个人做产品的来说，这是好消息。贵模型留给真正难的活，杂活交给便宜的，额度焦虑能少一大半。

https://t.co/HIxL6HhyYl

[Russell Kaplan @russelljkaplan]

This is directionally consistent with our internal data @cognition. Many tasks within software engineering are now "intelligence saturated." Once any model can do the task well, cost and speed are all that matter.