🏗️ (What Comes) After FOSS?
Framing open source as infrastructure rather than merely software shifts attention toward long-term maintenance, governance, and funding, especially for critical components that society depends on but that may lack a sustainable commercial model.
The entry describes “Open Digital Infrastructure” as the shared body of open-source code, standards, and knowledge underpinning software libraries, compilers, network protocols, and other digital building blocks. These resources are created by volunteer communities, research institutions, small businesses, and corporations. Together, they form a public technical foundation that supports both programming and essential functions across society.
🔗 Read more 🔗
🔐 Passphrase-less reboots using kexec under NixOS
The design improves unattended reboot reliability without storing a permanent decryption secret where the boot process can access it. It also shows how kexec can use volatile system state as a controlled bridge between kernel instances while bypassing slower firmware and bootloader stages.
Bevuta describes a NixOS technique for rebooting servers with encrypted disks without requiring an administrator to manually enter the LUKS passphrase. The approach uses Linux kexec to replace the running kernel without a full hardware restart and transfers a temporary decryption credential through volatile RAM. The implementation creates a one-time LUKS passphrase in a temporary keyslot, removes that keyslot after a successful kexec, and avoids exposing the credential on the kernel command line.
🔗 Read more 🔗
📝 AI;DR (AI; Didn’t Read)
The piece captures an emerging norm in AI-assisted communication: the concern is increasingly not whether AI was used, but whether the sender applied enough judgment to make the result worth someone else’s attention.
Rick Manelius argues for an “AI;DR” convention: if someone sends or publishes raw, unedited AI-generated prose, readers should feel free not to read it. He accepts AI as a normal part of the writing workflow but distinguishes between using it as an aid and putting unreviewed model output directly in front of colleagues or audiences. Human editing and ownership of the final prose, he suggests, remain important signals of care and respect.
🔗 Read more 🔗
💳 Stripe will reportedly acquire OpenRouter for $7B+
If completed at the reported price, the acquisition would represent a dramatic valuation increase in just a few months and give Stripe a strategic position in the infrastructure layer that brokers access across competing AI model providers.
TechCrunch reports, citing Bloomberg, that Stripe has finalized a deal to acquire AI model-routing startup OpenRouter for more than $7 billion. OpenRouter provides a unified interface for accessing hundreds of AI models and announced a $113 million Series B at a reported $1.3 billion valuation in May. Stripe declined to confirm the report, saying it does not comment on rumors or speculation.
🔗 Read more 🔗
🧠 Qwen 3.8 27B is excellent, but it defaults to overthinking things
A strong model in the 27B class pushes capable multimodal AI further into hardware that individuals can realistically own, although overly aggressive reasoning can introduce unnecessary latency and compute costs.
Simon Willison highlights Alibaba’s Qwen 3.8 27B, an Apache 2-licensed, vision-capable language model with 27 billion parameters. He considers it particularly attractive for local use on reasonably powerful laptops and notes that Qwen’s reported benchmarks show an improvement over its predecessor. His main reservation is that the model defaults to excessive reasoning, or “overthinking.”
🔗 Read more 🔗
📚 We Tracked a Shipment of Rare Books. It Ended at an Amazon AI Training Facility
The AirTag experiment provides unusually tangible evidence linking opaque bulk book acquisitions to industrial-scale digitization, sharpening questions about how AI companies obtain physical texts for training datasets and what happens to those books afterward.
Simon Willison summarizes a 404 Media investigation into mysterious bulk book purchases suspected of supplying AI training operations. A bookseller placed an AirTag supplied by the reporters inside an order of roughly 1,000 books, and the tracked book ultimately arrived at an Amazon facility in Las Vegas. Discussions among Amazon workers reportedly indicated that the facility destructively scans large quantities of books.
🔗 Read more 🔗