May the bridges we burn light the way.

I spent some time off reading some papers and thought that I could improve upon what was previously the world's fastest CRDT Introducing eg-strider: a set of improvements on top of Eg-walker that result in significantly better performance on every measurable axis jzhao.xyz/posts/eg-strider github.com/jackyzha0/eg-stri… (now featuring exclusive twitter language balls benchmark visualizations)

Vector has redeveloped its IEEE 11073 SDC stack in Rust, for medical device networking. 🦀 SDC, Service-oriented Device Connectivity, is a family of standards within IEEE 11073. It lets point-of-care medical devices from different vendors discover each other, exchange measurements, states and alarms, and expose their functions as standardized services. Vector unveiled what it calls the first Rust-based SDC stack at its Medical Engineering Symposium on October 15. It is expected to be available in the first quarter of 2027. Why Rust, as Vector puts it: → Memory and type safety help prevent common memory-handling errors during development → That can reduce certain causes of runtime errors and memory-related security vulnerabilities → For safety-critical devices, it is described as a solid foundation for reliability and cybersecurity requirements On integration: → The stack can be used from C and C++ applications through a foreign function interface → Manufacturers do not need to port existing application software to Rust → The architecture is designed to work across different hardware and operating systems On performance, Vector claims significantly better results, credited to optimized XML message processing and more efficient management of the device data structure. No numbers were given. A few things worth noting: → Vector's press release does not mention any certification or regulatory approval. One trade-press headline says the stack "qualifies for medical devices", but its text does not name a qualification either → The stack is not released yet → All claims are the vendor's own. No benchmarks, customers or toolchain details were given Still, a notable signal: a long-established automotive and medical tooling vendor is choosing Rust for a protocol stack intended to run inside medical devices. 🔗 vector.com/en/company/press/… #RustLang #MedTech #Embedded #Safety

How to find ALL PHP bugs: 1. Compile code (e.g. to RISC-V, x86, LLVM IR) 2. Vibe prove ASAN and non-ASAN versions have same behavior across all inputs. Hard, but Claude can do it! 3. Find a LOT of bugs Super proud of our work. A big step towards finding ALL memory safety bugs

The most vulnerable companies in software today are those selling pure convenience. If you ship an abstraction on top of AWS, Azure, GCP, or an OSS database... you are vulnerable. You should be actively deploying capital to get off the hyperscaler needle (be brave) and funding research to ship a genuinely novel product. I'd go as far as saying you're already seeing it on the books. AI is going to remove the willingness to pay for that abstraction.

I see it. I can see it. The vision is there. - Data Infra via the plane - Dev Infra via the ADE - Compute/Hardware Infra via campuses - Manufacturing via grit/determination - SMR Power via lobbying/determination

I don’t really believe in coincidence when it hits over and over and over. There is something going on with these vulnerabilities. I’m starting to feel like it’s not the thing we should be worried about, but rather a distraction to the thing we should be worried about.

The obvious lesson is CA certificates need to leave the Stone Age and move to threshold ECDSA signatures. The root programs need a blanket policy of only accepting certs with secret sharing between multiple independent HSMs. Crypto custodians figured this out years ago. When you have really important private keys that require resistance to nation state actors, the only solution is combining multiple independent sources of entropy

What is the best data room tooling? Are we building this ourselves? No way, right?

1. What does "Code Is Cheap" mean? How does it interact with open-source projects? 2. When code is cheap, is open-source (collaboration) still necessary? 3. What can we do to better maintain open-source projects today?

There are ≈3.1 million roots spread across ≈350k crates to test against on crates.io. It’s an absolute gift to Rust tooling development.

I think like 90%, conservatively, of the companies that exist in tech today are dog shit and will die in the next few years.

Unfinished coding side projects are now way easier to complete than unfinished blog posts. I’m perfectly OK with AI writing my code, even the docs. But I flinch at the idea of AI writing my blog posts, which feels like an egregious violation of a fundamental ethnical contract.

Every single Eng leader out there is iterating on some kind of PR autoapprover, coding automations, AI cost optimization, and model switching/eval stuff. Just a million people doing the same thing.

Wait a second. Planetscale runs on fucking Vercel?

I need an angel that wants to change the world. My tech is validated. I need to hire a few people. I need to sharpen the UX and infra.

I have a really extensive, wildly challenging test suite for validating my data infra against CXL and RDMA. I mean, I’ve been working on it and thinking about it everyday for two years. Today, it is green and I’m using no coherence protocol or locks.

I cannot believe OpenAI have fallen so badly. It's not even close. I'm using Astra Max and 6.1 Max... it's literally useless. It cannot do serious work.

Opus 5.5 is a really good model, but the security flags are awful. They're just not at all genuine. They're almost 'keyword' bad.

What does everyone use to manage your VPC connections, prebaked images, caches, remote connection commands, tooling updates, etc. Are we all manually managing this?