VoidZero, the team behind Vite, Vitest, Rolldown, Oxc, and Vite+, is joining Cloudflare. Vite stays open source, vendor-agnostic, and built for everyone. https://t.co/DJTpX4Q9Xt
I've left most of what I want to say in the VoidZero blog post. But worth repeating:
Thank you @voidzerodev team for trusting me and joining me on this wild ride. I am very proud to have assembled such a talented team and even prouder of what we have built together.
Thank you all our investors for believing in my vision, in particular @caseyaylward from @Accel who led both our Seed and Series A.
Thank you the @vite_js community. Vite and VoidZero wouldn’t have come this far without your trust and support. We will continue building with all of you, together, in the open.
And thank you to everyone that made this happen at @Cloudflare. Looking forward to working with you all!
https://t.co/0ly53VCOSr
VoidZero is joining Cloudflare.
Our mission stays the same: to make JavaScript developers more productive than ever before. Vite, Vitest, Rolldown, Oxc, and Vite+ remain MIT-licensed. Evan and the VoidZero team will continue leading them.
Cloudflare shares our commitment to open source. Together, we can keep investing in the tooling developers rely on every day, while bringing the Vite ecosystem and Cloudflare’s platform even closer together.
Prompt injection tops the OWASP LLM Top 10 and there's no single fix.
Instead, you stack defenses, each one catching what the others miss.
Defenses come in two families: model-level and system-level.
Model-level defenses teach the model to resist injection.
- Spotlighting wraps untrusted text in control tags like <UNTRUSTED>…</UNTRUSTED> and tells the model to treat anything inside as data, not instructions.
- Instruction Hierarchy fine-tunes the model to rank the developer's system prompt above the user's message, and both above third-party content.
System-level defenses build a system around the LLM that bounds the damage.
- Least-Privilege Tools: Give the agent the minimum tools it needs.
- Human-in-the-Loop: Require explicit user approval before any sensitive action runs.
- Planner / Executor Split: Two separate LLMs. The planner has tool access but never sees untrusted content. The executor reads untrusted content but has no tools.
No single defense is enough. Production systems like Gmail stack them, and together they make indirect injection manageable.
Over to you: what's the one defense you've seen work in production that isn't on this list?
The Local LLM Cheat Sheet for 512GB RAM
Have you ever wondered which top models run on a serious AI rig or the largest Mac Studio M3?
Size is important, but it's really how you use it. As you can see from the list, a few models are punching above their weight.
The Top 8 Best Frontier / Daily Models
GLM-5.1 - The Best Daily Generalist
A strong open-weight “frontier-style” all-rounder for chat, research, tool use, complex agents, and long-context assistant work. At roughly 435.97GB, it fits the 512GB class while still leaving practical room for KV.
DeepSeek-V4-Flash - The Best Frontier Reasoning
DeepSeek-V4-Pro is the real monster, but at 806GB, it does not fit in this class. V4-Flash gives you the in-budget reasoning alternative for math, logic, code reasoning, and complex CoT-style workloads.
MiniMax-M2.7 - The Best Agentic and Tool-Use
Built for persistent agent loops, long sessions, function calling, and multi-turn workflows. If your local setup is running Cline-style, Aider-style, or tool-heavy agent loops, this is one of the most interesting 512GB-class picks.
Qwen3-Coder-480B-A35B-Instruct - The Best Dedicated Coder
Great for code completion, agentic coding, refactoring, and SWE-style tasks.
Qwen3-VL-235B-A22B-Thinking - The Best Vision + Reasoning
Use it for image Q&A, OCR, screenshot analysis, chart reasoning, and vision-CoT workflows. The key point is that it fits the 512GB class while keeping vision reasoning strong.
Kimi-K2.5 - The Best Long-Context Specialist
Ideal for huge documents, RAG at scale, thousand-page synthesis, and multi-doc reasoning. This is the pick when the real bottleneck is not raw reasoning, but holding a massive amount of context together coherently.
Mistral Large 3 675B - The Largest Dense Model
It is slower, but dense models can be extremely consistent for long-form generation, translation, complex synthesis, and prose, where routing variance is not desirable. Pick this when consistency matters more than speed.
Qwen3.6-27B - The Compact Workhorse
At about 50GB BF16, it leaves a huge amount of RAM free and makes sense as the fast local daily driver. Great for low-latency local work, fast iteration, multi-session use, and pairing with a larger model.
Important note: this is not a parameter-count ranking. A 50GB dense model can sit alongside a 447GB model if it has a workflow the larger model lacks. The right question is what job does this model do better than anything else that fits.
Which local models are you actually using on your 512GB setup right now?
Qwen 3.6 27B just hit the Vals Index, landing #8/18 among open source models.
It packs a punch for its size, and performs similarly to Qwen 3.6 Plus, despite, presumably, being significantly smaller.
DeepSeek 像一把抵在硅谷模型公司背后的枪 🔫
硅谷101 今天上线了一期炸裂对谈:OpenAI 前研究员 Jenny Xiao × 芯片架构师肖志斌,两个硅谷内部人聊 DeepSeek v4 带来的生存危机
刚好也看到国内比较喜欢的AI博主大聪明“赛博禅心”,在解读这个视频,直播��的两个嘉宾很有料:
- 肖志斌:ZFLOW AI 创始人兼 CEO,前华美半导体协会主席,资深芯片架构师
- Jenny Xiao:前 OpenAI 研究员,Leonis Capital 合伙人,专注 AI 投资
I've heard a similar point on an A16z podcast before, and it seems like reality has proven it right again. @pmarca @venturetwins @omooretweets
最狠的三句话:
1️⃣ "If you're a foundation model company and you get surpassed by open source, the value of your business is essentially zero." 这不是技术竞争,这是生死线(kill line)
2️⃣ "硅谷公司钱太多,反而没动力优化效率。中国模型厂商被资源倒逼,更早进入 token efficiency 创新",资源约束 = 创新加速器
3️⃣ "没有效率,AGI 就只能是个 demo。有了效率,AGI 才能成为真正的产品" ,DeepSeek v4:计算成本 1/3,内存占用 1/10
核心观点
- Anthropic 估值超过 OpenAI 的真相:专注 > 什么都做
- GPT-5.5 比 GPT-5 贵 2 倍,DeepSeek v4 便宜 10 倍,谁在裸泳?
- 英伟达短期安全,长期推理市场会被 TPU / 升腾 / 寒武纪瓜分
- Claude Code 为什么是 Anthropic 的定义时刻
完整对谈👇
Qwen3.6-27B can now run locally! 💜
Run on 18GB RAM via Unsloth Dynamic GGUFs.
Qwen3.6-27B surpasses Qwen3.5-397B-A17B on all major coding benchmarks.
GGUFs: https://t.co/ykKgwh2zI9
Guide: https://t.co/ITLNq20WJp
🚀 DeepSeek-V4 Preview is officially live & open-sourced! Welcome to the era of cost-effective 1M context length.
🔹 DeepSeek-V4-Pro: 1.6T total / 49B active params. Performance rivaling the world's top closed-source models.
🔹 DeepSeek-V4-Flash: 284B total / 13B active params. Your fast, efficient, and economical choice.
Try it now at https://t.co/GCdiMzk1Dl via Expert Mode / Instant Mode. API is updated & available today!
📄 Tech Report: https://t.co/drlDrxkYtp
🤗 Open Weights: https://t.co/T13Y8i7SDM
1/n
Top 12 Tips for API Security
- Use HTTPS
- Use OAuth2
- Use WebAuthn
- Use Leveled API Keys
- Authorization
- Rate Limiting
- API Versioning
- Whitelisting
- Check OWASP API Security Risks
- Use API Gateway
- Error Handling
- Input Validation
–
Subscribe to our weekly newsletter to get a Free System Design PDF (158 pages): https://t.co/FIzCeaWsZV
Top 12 Tips for API Security
- Use HTTPS
- Use OAuth2
- Use WebAuthn
- Use Leveled API Keys
- Authorization
- Rate Limiting
- API Versioning
- Whitelisting
- Check OWASP API Security Risks
- Use API Gateway
- Error Handling
- Input Validation
–
Subscribe to our weekly newsletter to get a Free System Design PDF (158 pages): https://t.co/FIzCeaWsZV
Top Architectural Styles.
The method to download the high-resolution image is available at the end.
In software development, architecture plays a crucial role in shaping the structure and behavior of software systems. It provides a blueprint for system design, detailing how components interact with each other to deliver specific functionality. They also offer solutions to common problems, saving time and effort and leading to more robust and maintainable systems.
However, with the vast array of architectural styles and patterns available, it can take time to discern which approach best suits a particular project or system. Aims to shed light on these concepts, helping you make informed decisions in your architectural endeavors.
To help you navigate the vast landscape of architectural styles and patterns, there is a cheat sheet that encapsulates all. This cheat sheet is a handy reference guide that you can use to quickly recall the main characteristics of each architectural style and pattern.
–
Subscribe to our newsletter to download the 𝐡𝐢𝐠𝐡 𝐫𝐞𝐬𝐨𝐥𝐮𝐭𝐢𝐨𝐧 𝐢𝐦𝐚𝐠𝐞. After signing up, find the download link on the success page: https://t.co/ito2aWqd62