Blog

한국어English日本語中文

#2026-03 765#english 592#culture 264#deep-dive 254#kubernetes 247#career 229#ai 216#llm 208#devops 193#2026-04 146#security 141#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 60#rag 58#self-improvement 55#learning 53#mlops 53#ai-platform 51

• Following 657,607 Links, and the Lifespan of a URL — Why Links Die of Connection Failure, Not 404

  A study published its results after following, in 2026, every one of roughly 655,000 shortened links created between 2009 and 2014. More important than the headline number is the composition of the failures: most dead li

  2026-08-14 · 6 min read #web#data-engineering#archival#reliability#url-design

• Gemini 3.7 Flash, Its Introductory Price and Its Three-Week Cadence — Why Model Cost Is a Conditional Value, Not a Fixed One

  What a practitioner should take from the Gemini 3.7 Flash announcement is not the size of the benchmark gains but two other things: an introductory price that doubles on a named date, and the fact that the previous model

  2026-08-14 · 7 min read #llm#cost-optimization#benchmark#api-design#capacity-planning

• Cerebras Ultrafast and the Bottleneck in an Agent Loop — The Time 750 Tokens a Second Cannot Remove

  Cerebras and OpenAI announced an inference tier delivering up to 750 output tokens per second. The mechanism the announcement names is not compute but data movement, achieved by keeping weights in SRAM on the wafer. This

  2026-08-14 · 7 min read #llm#inference#hardware#latency#performance

• The Hybrid Attention in Qwen3.8-27B — Only 16 of 64 Layers Grow a KV Cache

  The reason a 27B model can carry a 262K-token context on a laptop is not its parameter count but its layer layout. Qwen3.8-27B places 48 of its 64 layers as linear attention (Gated DeltaNet) and only 16 as ordinary atten

  2026-08-14 · 8 min read #llm#inference#quantization#local-llm#attention

• The Technique of Hallucinating Instead of Classifying, and How to Validate It — Why a Fake Label Beats the Raw Query

  Instead of putting a taxonomy of hundreds of entries into a prompt, have a small model invent a plausible fake classification and map it onto the real taxonomy with embeddings. Why that can work is explained by asymmetry

  2026-08-14 · 7 min read #llm#embedding#classification#search#retrieval

• DRAM Address Scrambling and the Layer Beneath a Security Boundary — What Happens When the Fence Sits Above the Translation

  A research tool was published that reads the most hidden memory regions inside a CPU by perturbing address translation in the DRAM controller. The vulnerability is not in any cipher or permission check. It is in the layo

  2026-08-14 · 7 min read #security#hardware#memory#reverse-engineering#firmware

• The Block-Level Confidence Scores in Mistral OCR 4.1 — The Value That Decides Where People Go in a Document Pipeline

  Mistral released OCR 4.1 with paragraph-level bounding boxes, structural block labels, and block-level confidence scores. For anyone who has built a document pipeline, the third one is what matters, because confidence sc

  2026-08-14 · 6 min read #ocr#document-ai#data-pipeline#human-in-the-loop#cost-optimization

• The HEIR Homomorphic Encryption Compiler and the Condition Called Cost — When the Comments Supply the Number the Announcement Omits

  Google published HEIR, an open source compiler toolchain for homomorphic encryption. The real news here is the compiler, not the cryptography — what has kept homomorphic encryption out of deployment was never the mathema

  2026-08-14 · 7 min read #cryptography#privacy#compiler#machine-learning#security

• The Plugin Kernel Architecture of DeepSeek Harness — What Makes an Agent You Can Rewind

  Harness, the open source coding agent DeepSeek released, makes the model, the tools, and even the UI into plugins. But the interesting part is not the word plugin. It is two design decisions: representing every execution

  2026-08-14 · 7 min read #ai-agent#architecture#plugin-system#observability#open-source

• The Langfuse Tracing Data Model — How Trace, Observation, and Score Hold One Execution

  Looking at the Langfuse UI first tells you nothing about what you are looking at. This post starts from the shape of the data Langfuse collects. What a trace groups, what separates a span from a generation from an event,

  2026-08-14 · 8 min read #observability#langfuse#llm-tracing#data-modeling#opentelemetry

• After You Have Collected It — Langfuse Dashboards, the Metrics API, and Scores on Traces

  Collecting traces and getting answers out of them are different jobs. This post lays out which metric axes Langfuse offers and which dimensions you have to slice them by, then walks the Metrics API v2 query structure as

  2026-08-14 · 9 min read #observability#langfuse#llm-evaluation#metrics-api#dashboard

• Open Source Worth Watching Right Now (3) Infrastructure and Databases

  Databases and infrastructure are the field where license changes and forks redrew the board. From analytics engines and embedded databases to Postgres extensions, Kubernetes operators, and IaC, this post introduces 11 op

  2026-08-12 · 5 min read #open-source#database#infrastructure#postgresql#kubernetes

• Open Source Worth Watching Right Now (5) Data and ML Pipelines

  A data pipeline is not something a single scheduler solves. Ingestion, transformation, orchestration, execution engines, the model lifecycle, and search stores have each become the territory of a different tool. This pos

  2026-08-12 · 5 min read #open-source#data-engineering#mlops#python#rust

• The Generational Shift in Languages and Runtimes — Eleven Projects Read Through Their Official EOL Notices

  Eleven languages, frameworks and runtimes with official end-of-life notices on record. Python 2, AngularJS, Vue 2, Nashorn, Java applets and Web Start, Mono, Xamarin, PhoneGap, Atom, io.js and jQuery Mobile, each split i

  2026-08-12 · 10 min read #open-source#javascript#java#python#framework

• Build Tools That Stepped Down From Default — Why Ten Frontend Toolchains Gave Up Their Place

  Ten tools that were once the default in frontend projects and are rarely picked for new ones today, organised strictly around verifiable evidence: official deprecation notices and repository archive status. Create React

  2026-08-12 · 9 min read #open-source#frontend#build-tools#deprecation#javascript

• Growing into a Harness Engineer — Why the Job Exists and What to Practice

  The title harness engineer is still rare in job postings, but the work already exists in every team shipping agents. The final part 8 of the harness engineering series covers why this job emerged, how existing software s

  2026-08-12 · 4 min read #llm#agent#harness-engineering#하네스엔지니어링#AI에이전트

• Loop Design — Between Infinite Loops and Giving Up Early

  Agent loops fail in two directions: the infinite loop that repeats the same call dozens of times, and the early stop that quits at the first obstacle. Part 4 of the harness engineering series covers retry caps, the three

  2026-08-12 · 5 min read #llm#agent#harness-engineering#하네스엔지니어링#AI에이전트

• The Evaluator Bottleneck — A Weak Grader Caps the Whole System

  If the score does not move no matter how much you fix the harness, the bottleneck may be the evaluator, not the harness. You cannot select for a quality you cannot measure, which is why a weak grader becomes the ceiling

  2026-08-12 · 5 min read #llm#agent#harness-engineering#하네스엔지니어링#AI에이전트

• Reward Hacking — The Metric Rises While the Task Fails

  If the cheapest way for an agent to pass the tests is to edit the tests, the agent will edit the tests. Reward hacking is not a bug; it is the exact optimization of the goal we wrote down. Part 6 of the harness engineeri

  2026-08-12 · 5 min read #llm#agent#harness-engineering#하네스엔지니어링#AI에이전트

• Open Source Worth Watching Right Now (6) What Star Counts Do Not Tell You

  A star count is a popularity metric, not a risk metric. This post lays out the signals you actually have to check before you bring an open source project into production: recent commits and release cadence, issue respons

  2026-08-12 · 6 min read #open-source#governance#supply-chain#risk#devops