Blog
Writing on GPUs, LLMs, MLOps, Kubernetes — and mindset · 3525 posts
#2026-03 765#english 592#culture 264#deep-dive 254#kubernetes 249#career 229#ai 219#llm 209#devops 196#2026-04 146#security 143#database 114#observability 113#communication 109#history 107#architecture 100#productivity 96#finance 88#economy 84#mindset 81#psychology 80#ai-papers 79#food 78#it 78#travel 78#deep-learning 77#japanese 77#networking 77#performance 72#business-travel 70#linux 70#gpu 69#ai-agent 66#cs-fundamentals 63#postgresql 61#rag 58#self-improvement 55#learning 53#mlops 53#python 52
Talos Linux 1.13 — How a Shell-less Immutable OS Ended Up Shipping a Debug Shell
Talos Linux, the Kubernetes-only immutable OS, shipped 1.13 on April 27, 2026. The release brings a kernel built with Clang/ThinLTO, reproducible disk images, machine-wide container image signature verification, and its
2026-07-17 · 13 min read #linux#kubernetes#talos#immutable-infrastructure#devopsRAG · Fine-Tuning · Long-Context — Which One for My Problem: What the Papers Actually Measured, and What No One Measured ♪ Listenable
It's the most common question in LLM architecture, yet most answers are decision trees with no sources. This post answers only with what has been measured. That fine-tuning fails at injecting new knowledge has been measu
2026-07-17 · 23 min read #rag#llm#fine-tuning#long-context#aiPodman 6.0 — slirp4netns, CNI, and cgroups v1 Removed; macOS's Default Provider Is Now libkrun
On June 24, 2026, Podman 6.0.0 shipped as its first major release in 27 months since 5.0. Support for slirp4netns, CNI, iptables, cgroups v1, and BoltDB was removed all at once, Intel Mac and Windows 10 support was dropp
2026-07-17 · 13 min read #container#podman#docker#linux#macosA Mid-Course Check on Neo4j's CalVer Era: Block Format, Cypher 25, and How GQL Lands in the Engine
Neo4j switched to calendar versioning in early 2025, and over the roughly 17 months since then has shipped 18 feature releases — with each release's hotfix support ending the moment the next one ships. As of this writing
2026-07-17 · 14 min read #database#neo4j#graph-database#cypher#gqlTemporal Worker Versioning GA — The Deploy Problem Replay Created, and the Third Answer After Two Discarded Generations
Durable execution engines like Temporal survive failures by replaying an event history — and it is precisely that replay which makes shipping new code while workflows are still running the hardest problem in this model.
2026-07-17 · 13 min read #distributed-systems#durable-execution#temporal#workflow-engine#reliabilityLabel Locality Scheduling in Ray 2.56 — Placement Groups Start Seeing NVLink Racks, Not Nodes
Ray 2.56.0, released on June 29, 2026, adds a domain-level scheduling layer to placement groups as an alpha feature. Until now every placement strategy — PACK, STRICTPACK and friends — operated strictly at node granulari
2026-07-17 · 11 min read #ray#gpu#scheduling#distributed-systemsWhich Release Shipped Wayland Color Management — The Five-Year Deployment Timeline of a Protocol, and the X Server in 2026
Wayland's color-management and HDR protocol, color-management-v1, is the product of a five-year standardization effort — a merge request opened in January 2020 that was not merged until February 2025. This post pins down
2026-07-17 · 14 min read #linux#wayland#hdr#color-management#x11How Far Has OpenBao Actually Diverged from Vault — Tracing the Fork Through Release Notes, Up to v2.6's Namespace Sealing
Exactly two years have passed since OpenBao's v2.0.0 GA (2024-07-17), and v2.6.0 shipped three days ago. This post sets aside the tribal framing — 'BUSL betrayal' versus 'sustainability' — and checks where the fork actua
2026-07-17 · 10 min read #security#open-source#devops#licensingServo Embedding Landed on crates.io While Verso Stalled — A 2026 Mid-Year Check on Webview Fragmentation
Tauri keeps its binaries small by borrowing the OS webview (WebKitGTK, WKWebView, WebView2), and the price is webview fragmentation — rendering that differs subtly across platforms. Verso, a Servo-based webview that drew
2026-07-17 · 12 min read #desktop#tauri#servo#webview#rustHow MinIO Community Edition Ended — A Timeline Reconstructed from the Repository, and Where Garage, SeaweedFS, and Ceph RGW Stand Now ♪ Listenable
On April 25, 2026 the minio/minio repository was archived on GitHub, and on July 14 the mc client repository went read-only too. There is no shortage of angry threads, but this post follows the record rather than the ang
2026-07-17 · 14 min read #storage#object-storage#s3#minio#open-sourceHow to Actually Cut Your LLM API Bill — Why "90% Off From Caching" Is 25% on the Invoice ♪ Listenable
A prompt cache read costs one-tenth of the input price. But that does not take 90% off your invoice. Follow the worked example Anthropic publishes in its own docs and, even with the cache fully warm, the total drops only
2026-07-17 · 22 min read #llm#cost-optimization#prompt-caching#api#aiHome Assistant Matter Server 9.0 — Why It Dropped the Official C++ SDK for a matter.js Rewrite
On June 23, 2026, Home Assistant's Matter server jumped to 9.0, and with it the Python plus official C++ SDK (connectedhomeip) implementation was replaced wholesale by a new server built on matter.js, rewritten from scra
2026-07-17 · 11 min read #iot#smart-home#home-assistant#matter#threadReading OpenSSH 10.4 Closely — Experimental ML-DSA Composite Signatures, the Next Step After the mlkem768 Default, and the Deprecations That Bite Operators
OpenSSH 10.4, released on July 6, 2026, looks like a bugfix release on the surface, but read the release notes closely and two threads emerge. One is the next step of the post-quantum transition — key exchange was effect
2026-07-17 · 15 min read #security#openssh#ssh#post-quantum-cryptography#cryptographyThe State of Media over QUIC — Spec at draft-19, Deployment at draft-07, and Its First Author Has Forked to moq-lite
IETF Media over QUIC (MoQ) wants to merge WebRTC's interactive latency with HLS/DASH's scalability into a single architecture, and as of July 6, 2026 it has reached draft-ietf-moq-transport-19, with the working group's m
2026-07-17 · 13 min read #network#quic#streaming#webrtc#rfctorchcomms in PyTorch 2.13 — The New Distributed Communication Backend Coming for c10d
Among the highlights of PyTorch 2.13, released on July 8, 2026, the most structural change is torchcomms — PyTorch Distributed's new communication backend has begun landing in core's CI and DeviceMesh paths. torchcomms i
2026-07-17 · 10 min read #pytorch#distributed-training#nccl#deep-learningNATS Server 2.14 — Fast Batch Publish, Built-In Server Scheduling, and the Homework Due Before 2.15
NATS Server 2.14, released on April 30, 2026, adds fast batch publish — high-throughput batch publishing to JetStream with built-in flow control — cron-style recurring schedules with subject sampling, automatic conversio
2026-07-17 · 14 min read #messaging#nats#distributed-systems#release-notesHow Real Is the JPEG XL Comeback — 8 Months After Chromium's Reversal, Verified From the Source
In October 2022, Chromium removed the JPEG XL code, citing a lack of ecosystem interest, and that decision stayed the go-to example in image-format debates for over three years. On November 22, 2025, Chrome's ATLs revers
2026-07-17 · 13 min read #multimedia#jpeg-xl#chromium#browser#web-performanceDMARC Becomes a Standard After 11 Years — RFC 9989's Tree Walk, the pct Sunset, and the p=reject Warning
DMARC (RFC 7489), the de facto standard for email authentication since March 2015, was actually an Informational document that never went through IETF working-group consensus. In May 2026 the IETF replaced it with three
2026-07-17 · 15 min read #email#dmarc#dns#security#rfcHow Much VRAM to Run an LLM Locally — Compute It From Formulas, Not Tables ♪ Listenable
The answer to "how many GB does an 8B model need?" is not a table but two formulas. Weights are parameter count times bpw divided by 8; the KV cache is 2 times layers times KV heads times headdim times bytes times tokens
2026-07-17 · 24 min read #llm#quantization#kv-cache#local-llm#gpuFlux 2.9 and Two Years After Weaveworks — How a GitOps Project Survives Losing Its Sponsor
When Weaveworks, the company that created Flux, shut down in early 2024, a GitHub discussion appeared asking: is the project future at risk? Two and a half years later, on 2026-06-30, Flux shipped 2.9.0. This post recons
2026-07-17 · 12 min read #devops#gitops#fluxcd#kubernetes#open-source