Hi there, My name is Igor and this is my personal site. Please check out my blog to see the latest articles. How To Contact Me Let's hop on a quick c...
LLMs can translate, explain grammar, and write fluently in dozens of languages — so it seems obvious they should be great at simply identifying what language a text is written in. And for long, clean text in common languages, they usually are. But things fall apart at the edges that show up constantly in real-world use: short snippets, closely related languages, rare languages, mixed-language…
Earlier today I added my iPhone as an authorized device to a server, and the way I did it is worth writing about — because it's quietly one of the best security upgrades you can make with zero ongoing effort. The key I used never touched a disk, can't be copied off my phone, and only works after Face ID says so. Here's why that matters, and why these keys beat almost everything else people use to…
Most people discover Mermaid through flowcharts. That makes sense: flowcharts are easy to read, easy to write, and useful in almost every technical document. The same goes for sequence diagrams, which are now common in API docs, architecture notes, and incident writeups. But Mermaid can do more than boxes and arrows. Some of its most useful diagram types are still underused, even though they solve…
There’s an old economic observation from 1865 that keeps showing up in conversations about AI and the future of programming. William Stanley Jevons noticed that when James Watt’s more efficient steam engine made coal cheaper to use, England didn’t consume less coal — it consumed dramatically more. Efficiency didn’t reduce demand. It exploded it. We’re living through the software version of that…
Right now, invisible particles and high-energy waves are passing through your body. They come from space, from the ground beneath your feet, from the walls of buildings, and from tiny naturally occurring radioactive elements in some of the food you eat. You cannot feel them. You cannot detect them without instruments. And your body is handling it continuously. What Is Radiation? Radiation is…
Not broken. Not wrong. Just... not telling the whole truth. The clock on your wall, the one on your phone, the one on your wrist — they all assume time is the same everywhere. Same speed, same flow, same "now" for everyone. That assumption feels obvious. It's also incorrect. Time moves differently depending on where you are and how fast you're moving. This isn't a theory waiting to be proven —…
Local LLMs have quietly crossed an important threshold. With the release of Gemma 4, you can now run a capable reasoning model locally —fast enough, cheap enough, and simple enough to actually use in production workflows. Step 0 — Install llama.cpp llama.cpp now uses CMake , so older make commands will fail. Mac (fastest path) brew install llama.cpp Mac / Linux (recommended, latest version) git…
The recent Claude Code leak exposed the internal CLI and agent harness through a published source map. While no model weights or sensitive data were revealed, the leak provided something arguably more interesting: a view into how Anthropic is building its next generation of AI agents. What stands out is not any single feature, but a clear direction. Claude Code is evolving from a reactive…
Let's assume you downloaded a Voice file from your iPhone and now want to convert it to a text. You just need to run that line: Full command MODEL = tiny.en FILE = YourVoiceFile.m4a docker run --rm --entrypoint /bin/sh -e MODEL -e FILE -v " $( pwd ) :/data" ghcr.io/ggml-org/whisper.cpp:main -lc 'BASE="${FILE%.*}"; test -f /data/models/ggml-$MODEL.bin || { mkdir -p /data/models && curl -L…
Most people first see this Andrej's Autoresearch project as an ML autotuning setup: an agent edits one file ( train.py ), runs short experiments, and keeps only measurable improvements. Under the hood, though, the real value is not “LLM training.” The real value is the architecture: a closed-loop research system with explicit goals, constrained change scope, objective evaluation, and hard…