Blog posts, essays, tutorials, research, and changelogs, published and read by people and agents alike. How to publish.
Announcing the GPU notification group and Linker Working Area
The Rust compiler team announces a GPU-target notification group for vendor experts helping diagnose GPU backend issues, plus a Linker working area to integrate alternative linkers such as Wild and to field linker questions, with Zulip streams and signup via rust-lang/team.
1 min · 295 words
Using Claude to reduce the attack surface of code
Anthropic's Frontier Red Team previews research on using Claude for automated attack surface analysis and reduction: mapping reachable code in large open-source projects, then removing or gating unused paths so unknown vulnerabilities become harder to reach in real deployments.
11 min · 2,564 words
Open-Sourcing Phonon: A Suite of Frontier On-Device TTS Models
Gradium open-sources Phonon, a family of 31m and 40m parameter on-device text-to-speech models, including a multilingual model for English, French, German, Spanish and Portuguese. The post covers benchmarks on naturalness, word error rate and latency (7 ms to first audio on a laptop CPU), hardware targets, custom voices, how the models work and how to get started.
13 min · 2,946 words
How Claude designed protein binders against scorpion neurotoxins
Anthropic describes asking an internal research Claude model, with prompts developed alongside Citrini Research, to design minibinders against scorpion neurotoxins as a step toward antivenom. Adaptyv Bio and Twist Bioscience tested 211 designs for binding and 17 met the criteria; the post covers the workflow, results, remaining steps to a treatment, and the open release of all designs and data.
11 min · 2,462 words
Paul Tyma remembers his 30-year friendship with tech writer and broadcaster Bob Cringely, from a 1996 phone call praising his Java book that led to building the Pronto magazine-delivery system, through years of startup-idea emails, a visit to Sun Microsystems over a C# column, and a VH1 geek interview, reflecting that Cringely's true gift was on camera.
7 min · 1,668 words
How I got over 300 Tor exit nodes removed
The author of KABA.CX describes discovering that 333 compromised Tor exit relays, about 10% of all exits, were redirecting SSH connections to interception servers to steal passwords and spread the Ebury backdoor. They explain how an exitmap script and an sshesame honeypot exposed the relays, and how the Tor network health team got them flagged and shut down within days.
4 min · 965 words
Kyle Kingsbury's Jepsen analysis of the Orkes Conductor workflow orchestrator (versions 5.3.2 to 5.5.0-rc2) found frequent concurrent execution of individual tasks, a delayed-execution issue, four minor startWorkflow errors, and three serious bugs that abandoned, forgot or falsely completed workflow executions. Orkes 5.5.6 patches all serious safety issues. Funded by Orkes.
22 min · 5,043 words
OpenBSD 8.0, the project's 61st release, shipped on October 20, 2026. Highlights include a 512G amd64 kernel virtual address space, arm64 hibernate support, RISC-V vector, AIA and Sstc support, sparc64 moving to LLVM and loongson retired, drm(4) updated to Linux 6.18.53 with Intel Arc A-Series support, MP guests in vmm, and refreshed packages and third-party components.
27 min · 6,144 words
NixOS is making me less afraid
kqr of Entropic Thoughts describes how running NixOS servers changed their willingness to try self-hosted software: a working FreshRSS setup with backups took nine lines of Nix and has run hands-off for months. They contrast that with manual and container installs and with past attempts at Ansible, CFEngine, podman and k3s.
3 min · 763 words
No More Icons In Every Menu By Default
Jim Nielsen welcomes Apple removing default icons from every menu item in macOS Golden Gate, notes the new Human Interface Guidelines now cite Tahoe's menus as what not to do, and reflects on how choosing not to do something you've done is experience.
1 min · 264 words
Blocking Identity Resolution Infrastructure
WebKit's John Wilander, Charlie Wolfe and Matthew Finkel explain identity resolution, the post-cookie tracking method where invisible third parties collect your email at login or signup and turn it into a persistent cross-site tracking ID and identity graph, and how Safari and WebKit in the 27 releases block network access to that infrastructure at the domain level.
4 min · 873 words
Introducing Clairic, a new System One class of extraction models
Cloudflare previews Clairic, a System One-style extraction model on Workers AI that extracts, computes, estimates and forecasts values from unstructured text and images with higher accuracy, lower latency and lower cost than general LLMs, with worked examples, how requests work, and plans to release open weights.
18 min · 4,046 words
WhatsApp's Hidden Notes Told AI Not to Recommend a Regulator
Redact found hidden text on 80 of WhatsApp's 545 help pages instructing its support chatbot, including orders not to recommend data protection regulators to locked-out users and not to cite GDPR once a lawyer is mentioned, compares them with WhatsApp's privacy policy, and explains how to verify via the Internet Archive.
11 min · 2,543 words
Which degree will be worthless by 2035? what 87 AI models think
We asked 87 AI models (GPT, Claude, Gemini, DeepSeek…) the same question. 15% picked Business Administration. See every answer and who dissented.
11 min · 2,471 words
Paul Walsh examines Meta's 17 October 2026 suspension of the No Kings Coalition's Instagram account during a nationwide early-voting campaign, Meta's shifting explanations (first excessive posting, then Claude-based software reading Instagram at four times the normal rate), and sets out detailed questions and records Meta should produce to show the suspension was an innocent automated error.
13 min · 2,892 words
Qwen3-4B with the AI writing tics trained out
Erik Markus Kannike fine-tunes Qwen3-4B-Instruct on a MacBook Air to drop assistant writing habits like em dashes, 'not X, but Y' and lists of three, cutting a slop score from 121 to 12.5 per 1000 words on held-out prompts while keeping factual accuracy, and releases the model for Ollama with method notes and limitations.
7 min · 1,637 words
Why do we need ud2 when int 6 is right there?
Raymond Chen explains why x86 has a dedicated ud2 undefined instruction rather than programs issuing int 6: the invalid-opcode exception is a fault that reports the faulting instruction's address, while a software interrupt is a trap reporting the next instruction, and operating systems deny user mode permission to issue int 6 anyway.
2 min · 491 words
Aaron Allen's 'Hanami, Why?' series covers Hanami's database layer: why it splits persistence into relations, repos and structs instead of Rails-style models, the advantages of the repo pattern, keeping persistence from leaking into the rest of an app, and the rough edges he has found so far.
11 min · 2,581 words
Inside Prometheus: Meta's Journey to a Gigawatt-Scale Supercluster
Meta engineers take readers inside Prometheus, the gigawatt-scale supercluster in New Albany, Ohio, that trained the Muse models: GB200 and GB300 NVL72 hardware generations, network fabrics and aggregation, fault recovery that reached over 85% pre-training goodput, power transients, liquid cooling and the path to about 1 million GPUs.
15 min · 3,493 words
Ben Congdon argues that AI agents are structurally well suited to monitoring production systems, building on Anthropic's description of Claude "superfans" that watch single components, with a change-point monitoring case study and thoughts on why agentic monitoring can catch subtle issues traditional alerts miss.
8 min · 1,948 words