Skip to content

Loop Notes

  • AI Coding Agents
  • Backend & Servers
  • Tech in Korea
  • Hiring & Careers

KakaoTalk Verification Code Not Sent or Still in Progress?

November 17, 2026November 1, 2026 by jjeongil
Diagram of where a KakaoTalk verification code can get stuck: iPhone unknown senders filter, Samsung spam or blocked folder, or a carrier block

KakaoTalk code not coming? Stop at 3 tries. A foreign number can be asked to send a text; a Korean number on a new phone may have it in filtered messages.

Categories Tech in Korea

Does HackerRank Detect AI Cheating? An Interviewer’s View

November 3, 2026October 30, 2026 by jjeongil
Two-stage hiring funnel with automated coding assessment screening first and a live technical interview second

HackerRank logs copy-paste, tab switches, and code similarity, and sells AI-based checks. As an interviewer, I treat a flag as a reason to ask, not proof.

Categories Hiring & Careers

Claude Code or Codex Keeps Asking for Permission? 4 Causes

November 13, 2026October 29, 2026 by jjeongil
Diagram of Claude Code permission layers: deny and ask rules sit above the permission mode, and a deny rule blocks the Read tool on .env while a node -e script still reads it in bypass mode

Check the mode first; in Manual, switch to acceptEdits. Still asked? It’s usually your own ask rule, a && chain, an untrusted folder, or Codex’s sandbox.

Categories AI Coding Agents

Claude Pro vs Max: Why I Haven’t Upgraded for Claude Code

October 29, 2026 by jjeongil
Flowchart of four checks before upgrading from Claude Pro to Max: can it wait, can another tool take it, can overlapping subscriptions be cut, and does Claude still stop the work

I hit Claude Pro limits in Claude Code, but waiting and Codex cover the gap. I’d move to Max at 3+ delayed days or over 10 hours of waiting a month.

Categories AI Coding Agents

504 Gateway Timeout but No App Errors? Find Who Cut It Off

November 3, 2026October 28, 2026 by jjeongil
Diagram of a request going from client to front layer to Spring Boot app to downstream, with the front layer returning 502, 503, or 504 itself and a 504 at about 30 seconds while the app still waits

A 504 with clean app logs can mean a gateway or load balancer timeout ran out first. How I found mine near 30 seconds, and how I tell 502, 503, 504 apart.

Categories Backend & Servers

RTX 4090 Local LLM: Which Runtime, Model Size and Context

November 14, 2026October 27, 2026 by jjeongil
Diagram of three steps in order for a slow local LLM on an RTX 4090: run ollama ps and look for 100% GPU, check the context length, switch runtime last

On an RTX 4090, run Ollama with Qwen3-Coder 30B at 64K context. Slow? Check ollama ps first: of ten stalls I wrote up, switching runtimes fixed none.

Categories AI Coding Agents

STS “cannot be resolved to a type” but Gradle build passes

November 18, 2026October 27, 2026 by jjeongil
Diagram: a build.gradle dependency change reaches the terminal Gradle build, which passes, while the STS project model and classpath lag behind and show "cannot be resolved to a type" until Refresh Gradle Project

Terminal build passes but STS shows “cannot be resolved to a type”? Run Refresh Gradle Project, not Clean. Still red? The marker’s type tells you the fix.

Categories Backend & Servers

Mac Studio M3 Ultra 96GB vs 4090: Local LLM Review

November 5, 2026October 26, 2026 by jjeongil
Bar chart of Mac Studio M3 Ultra 96GB memory use for local LLMs: Llama 3.3 70B Q4, Qwen3-Coder 30B at 256K, both together, and Qwen3-235B Q2_K, against an estimated default GPU limit of about 72GB, the raised 88GB limit, and 96GB total

My Mac Studio M3 Ultra 96GB takes the 70B and 256K-context jobs my RTX 4090 can’t, but Qwen3-235B at Q2_K (85.7GB) broke a simple coding task.

Categories AI Coding Agents

Local LLM Too Slow? 8B vs 30B on My RTX 4090

November 5, 2026October 26, 2026 by jjeongil
Redrawn illustration of a terminal running a local LLM with the ollama run command, with the answer streaming in below

On my RTX 4090 (24GB), 7–8B models answer instantly; a 30B-class model took 20–30 s to start. Why big models slow down, how to check it, and what to pick.

Categories AI Coding Agents

How to Use Naver Map in English: A Korean Resident’s Guide

October 30, 2026October 25, 2026 by jjeongil
Naver Map desktop site in Korean showing no results for the romanized search "Starbucks Gangnam Station"

Naver Map’s desktop site was Korean-only in my October 2026 test, and romanized searches found 6 of 8 places, so here’s how I search as a Korean resident.

Categories Tech in Korea
Older posts
Newer posts
← Previous Page1 Page2 Page3 … Page5 Next →

Recent Posts

  • Claude Code Auto Mode Not Working: Check the Real Mode First
  • Ollama Not Using GPU After an Update on Linux: The Fix
  • Claude Code Ignores CLAUDE.md or Skills? Where to Look First
  • Local LLM Hardware: What to Buy for the Model You’ll Run
  • Claude Code Skill Not Triggering? It’s the Prompt (45 Runs)
  • About
  • Privacy Policy
  • Terms of Use
  • Editorial Policy
  • Contact
© 2026 Loop Notes