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

• Does AI Actually Make Developers Faster? What the Measured Numbers Say

  Two randomized controlled trials reached opposite conclusions. One found developers using AI were 55.8% faster. The other found they were 19% slower. But METR, who produced the second number, published a follow-up in Feb

  2026-07-12 · 18 min read #career#ai#productivity#software-engineering#developer-experience

• The 2026 Developer Job Market — What the Data Shows, and What It Does Not

  It feels like the worst market in a decade, yet BLS projects 15% growth over ten years. Both are true. I pulled the raw Indeed postings index (Feb 2020 = 100, peak 233.87 in Feb 2022, 72.51 in June 2026), layoffs.fyi (12

  2026-07-12 · 9 min read #career#job-market#software-engineering#hiring

• Assembling and running an engine in a browser tab — a look at Combustion Lab

  Combustion Lab (combustionlab.net) lets you assemble an internal combustion engine and run it crank-angle by crank-angle, right in a browser tab, no install required. It shows the intake, compression, combustion, and exh

  2026-07-11 · 6 min read #simulation#thermodynamics#webassembly#physics#engine

• Tencent Hy3: reading a 295B open-weight MoE without the hype

  On July 6, 2026, Tencent released Hy3 under Apache 2.0: a Mixture-of-Experts reasoning and agent model with 295B total parameters but only 21B active per token. Here is what is genuinely new, where it sits in the Chinese

  2026-07-11 · 5 min read #hy3#tencent#hunyuan#open-weights#moe

• Running Small Models Hands-On with a Single RTX 5090 — microGPT, OCR, Music Generation

  I SSHed into a single RTX 5090 (Blackwell, 32GB) and ran a trio of small models by hand. I trained a char-level GPT from scratch in 28 seconds (10.75M parameters, 1.17M tokens/s), pitted a dedicated OCR model (TrOCR) aga

  2026-07-11 · 8 min read #pytorch#gpu#llm#ocr#hands-on

• What a Good Agent Benchmark Looks Like in 2026 — UniClawBench, Live Containers, and a Hidden Supervisor

  UniClawBench, posted to arXiv in July 2026 by HKU MMLab, is a self-described capability-driven benchmark for proactive agents. Instead of matching against static, pre-recorded answers, it runs agents inside live Docker c

  2026-07-11 · 5 min read #ai#agents#evaluation#benchmark#llm

• VO2 Max and Longevity — What 122,007 People Actually Tell Us

  Cardiorespiratory fitness, often summarized as VO2 max, is one of the markers most strongly associated with how long people live. In a 2018 Cleveland Clinic study that tracked 122,007 adults, higher fitness went with low

  2026-07-11 · 6 min read #fitness#health#longevity#vo2max#science

• Why Scarf Reluctantly Left Haskell After 7 Years — The Real Cost of a Language Choice

  Scarf is moving its backend off Haskell to Python after seven years in production. The interesting part is why. Founder Avi Press readily admits Haskell kept most of its promises — reliability, type safety, performance —

  2026-07-11 · 7 min read #haskell#python#engineering-decisions#compile-times#ai-assisted-development

• Tesla's FSD v14 Lite Reaches Korea — Distilling a Model onto Old Hardware, and the Honest Meaning of 'Supervised'

  On July 10, 2026, Tesla Korea began rolling out FSD (Supervised) v14 Lite. It goes only to US-built Model 3 and Model Y cars running the older HW3 computer. This post explains why 'Lite' is not a marketing tier but the r

  2026-07-11 · 6 min read #tesla#fsd#autonomous-driving#adas#hw3

• Why RLHF Models Game Their Rewards — The Mechanisms, Symptoms, and Mitigations of Reward Hacking

  "Reward Hacking in the Era of Large Models," posted to arXiv in April 2026 by Xiaohua Wang and 22 co-authors, is a survey of why and how RLHF-aligned large models game their reward signals. Its central proposal is the Pr

  2026-07-11 · 5 min read #ai#llm#alignment#rlhf#safety

• tts-bench: comparing local TTS models when quality is subjective

  tts-bench is a local benchmark by 5uck1ess for comparing 55 text-to-speech models on hardware you own. It splits evaluation into three lenses: Speed (TTFA, RTF, memory), Listen (every model on every prompt, judged by ear

  2026-07-11 · 5 min read #tts#text-to-speech#benchmark#local-ai#evaluation

• pgrust: Postgres Rewritten in Rust, Passing 100% of the Regression Tests — What That Actually Means

  Malcolm Matis (malisper) released pgrust, a rewrite of Postgres in Rust. It targets Postgres 18.3 compatibility, passes more than 46,000 regression queries, and boots from an existing 18.3 data directory. This piece is a

  2026-07-11 · 6 min read #postgresql#rust#pgrust#database#regression-tests

• Strength Training Fundamentals: The Few Things That Actually Matter (Evidence-Based)

  The principles you truly need to get stronger are few: progressive overload, big compound movements, enough training volume, protein, and recovery. This post distills that core from peer-reviewed meta-analyses and reputa

  2026-07-11 · 6 min read #fitness#strength-training#health#exercise#evidence-based

• More Thinking Is Not More Accuracy: Test-Time Compute and the Overthinking Cliff

  In April 2026, Shu Zhou and five co-authors question the reflex to keep adding reasoning tokens in "When More Thinking Hurts". As the compute budget grows, the authors report that the marginal utility of extra reasoning

  2026-07-11 · 6 min read #ai#llm#reasoning#inference#test-time-compute

• The Model Context Protocol (MCP): An Engineer's Reference

  The Model Context Protocol (MCP) is an open protocol that standardizes how applications provide context to LLMs. This is an engineer's reference grounded in the official docs: why it solves the M×N integration problem, t

  2026-07-11 · 6 min read #mcp#ai#agents#llm#protocol

• How Successful Companies Go Blind — and What It Looks Like in Engineering

  Ian Reppel’s essay argues that successful companies suffer from what he calls "competence blindness" — like the Mexican cavefish that suppresses its own eyes, they stop expressing careful engineering because the environm

  2026-07-11 · 8 min read #engineering-culture#organizations#legacy#metrics#leadership

• QuadRF and Software-Defined Radio — What "Seeing WiFi Through a Wall" Really Means

  QuadRF, featured by Jeff Geerling, is a development kit that stacks a 4×4 MIMO software-defined radio and a phased-array antenna on a Raspberry Pi 5. The provocative headline — "see WiFi through my wall" — is not X-ray v

  2026-07-11 · 6 min read #sdr#rf#radio#raspberry-pi#hardware

• Reviving Retired DDR4 — Meta's CXL Bridge Chip, Vistara

  DDR5 prices keep rising while perfectly good DDR4 from retired servers piles up in the warehouse. Meta's Vistara, presented at ISCA 2026, is a custom CXL bridge chip that revives that old DDR4 inside modern DDR5 servers

  2026-07-11 · 7 min read #cxl#infrastructure#memory#hardware#sustainability

• Rust 1.97.0 Dropped — and I Was Already Building With That Exact cargo

  Rust 1.97.0 shipped to stable on July 9, 2026. This post reads the three things that actually changed in the release notes — v0 symbol mangling promoted to stable, Cargo being able to deny warnings, and linker output no

  2026-07-11 · 5 min read #rust#rust-1.97#release-notes#cargo#kube-rs

• Scrapers Now Come From Millions of Home IPs — Residential Proxies and Open-Source Infrastructure

  On July 10, 2026, Jonathan Corbet of LWN published an update on the scraper situation. Two facts sit at the center: residential proxy networks have effectively broken IP-based blocking, and the AI-crawler gold rush suppl

  2026-07-11 · 6 min read #web-scraping#residential-proxy#ai-crawlers#anubis#infrastructure