Remote

Yurii / Юра 🇺🇦

@iurii@mastodon.world

mastodon 4.7.3

We need to write code… You son of a bitch, I’m in.

🇺🇦 Ukrainian. #SoftwareEngineering and #TechnicalLeadership. I also write about #politics and hobbies.

Music lover, all of it (lately #Metal & #Jazz). Amateur guitarist & vocalist. I visit #concerts and #festivals.

Vacation = Mountains. #Trekking or #Snowboarding — as long as it's high up.

0 Followers

0 Following

27 Posts

Joined January 28, 2026

Blog:

http://iurii.net

C/C++ cache:

https://cachelot.io

Golang brokerless pub/sub:

https://github.com/Aigent/nq

Yurii / Юра 🇺🇦 @iurii@mastodon.world

· 7mo ago

Replying to

@iris_meredith fuck, this hits hard. I would also include all the brilliant minds working hard on giving us depression from computers. I saw a random comment on Reddit: ... many smart people tie their inherent value to their professional contribution. I can relate. AuDHD, I live and sleep in a constant mild to high strees, it takes high effort to not escape discomfort from the real world and inside my head. SwE allows it plus the validation that I exist. But this validation is my output.

Yurii / Юра 🇺🇦 @iurii@mastodon.world

· 7mo ago

Replying to

@iris_meredith If I'd focus on one aspect only. I started to work in tech in 2000, IT crowd was in basements, away from real people, we were geeking out with a bunch of nerds, life was fun. Fast-forward, and billionaires like Thiel and Brin telling employers that 60 hours a week is the "sweet spot" and Thiel wannabes tell me to "inspire" team to work 60h/week also leading by example.

Yurii / Юра 🇺🇦 @iurii@mastodon.world

· 8mo ago

Replying to on mastodon.social

@Yoshi@toot.community @AdmSnackbar@mastodon.social @krystyna@toad.social @nixCraft@mastodon.social I cannot imagine how for profit companies would spend money on housing. We should tax ultra wealthy though, without hesitation. They hoard resources to the point they try to deprive the rest of us from political representation (Trump has already happened, and they financing corrupt far-right in EU and elsewhere)

Yurii / Юра 🇺🇦 @iurii@mastodon.world

· 8mo ago

Replying to

I used to work on a remotely hosted codebase, the only viable way to work was console editors. Almost all my colleagues used Vim, I thought it was the stupidest thing on Earth. To prove it, I learned Vim and made a few plugins to convincingly argue how stupid it is. I’ve been using Vim ever since. Not as much as an IDE, but like an ultimate file transformer and a quick editor. You take a raw, unstructured log dump and come back with a CSV file in 5 minutes, plus some regex scars.

Yurii / Юра 🇺🇦 @iurii@mastodon.world

· 8mo ago

I wish my parents told me, ‘Son, with enough dedication and passion, one day you could become a professional video game player.’

Yurii / Юра 🇺🇦 @iurii@mastodon.world

· 8mo ago

Every time you create a pointer to void, void creates a pointer to you.

Yurii / Юра 🇺🇦 @iurii@mastodon.world

· 8mo ago

Highly recommend using regex. For instance, here’s Tetris written with sed https://github.com/uuner/sedtris

GitHub

GitHub - uuner/sedtris: Tetris in sed

Tetris in sed. Contribute to uuner/sedtris development by creating an account on GitHub.

Yurii / Юра 🇺🇦 @iurii@mastodon.world

· 8mo ago

#Java is #javascript for communists with class struggle and factories for workers

Yurii / Юра 🇺🇦 @iurii@mastodon.world

· 8mo ago

Me at 25: I run custom, self-compiled Linux kernels only. Me at 39: My kid touched something on my phone, and now I don’t receive notifications.

Yurii / Юра 🇺🇦 @iurii@mastodon.world

· 7mo ago

This privacy essay is different from what I see online.

I often get this question "why do you care" or "what is the difference" so I structured the whole text the way I can easily reply with the link.

It tries to answer both, "WHY" and "HOW"

mainly, it's for the average person who doesn't know how or doesn't care enough to selectively enable webGL on certain websites or install Magisk, Zygisk, etc.

https://iurii.net/en/blog/posts/essays/privacy/

iurii.net

Yurii / Юра 🇺🇦 @iurii@mastodon.world

· 7mo ago

Look around! Autonomous agentic AIs running wild — opening PRs, deploying production.

But software engineering is a craft.

Conservative, meticulous multi-session vibe-coding beats this hipster agentic nonsense 10/10.

Yurii / Юра 🇺🇦 @iurii@mastodon.world

· 8mo ago

Replying to

@thedarktangent https://mastodon.world/@iurii/116018796456022288

mastodon.world

Yurii / Юра 🇺🇦: "@Powerfromspace1@mstdn.social Epstein has a lot o…" - Mastodon

Yurii / Юра 🇺🇦 @iurii@mastodon.world

· 5mo ago

Took me a second...

Yurii / Юра 🇺🇦 @iurii@mastodon.world

· 4mo ago

What do you mean Sleep Token's lore is intentionally ambiguous and full of gaps?

Two EPs and three albums are all about using Vim.

Opening first commit message unintentionally, getting seduced by the strange and seemingly beautiful form of interaction with such familiar text, trying to understand Vim better to reach that intimacy and become one with the editor, eventually realizing your code is getting scrambled while being unable to walk away, betrayed by your own muscle memory.

Yurii / Юра 🇺🇦 @iurii@mastodon.world

· 2mo ago

Replying to

@davidgerard@circumstances.run generally, even 4B-class 2 bit quantized LLMs are okay at tool calling. So if you provide a small model with a harness wrapping small set of verifiable tools, it performs okay. I'm trying to teach cybersecurity to 30B model, it performs good, provided you give it some time. Again, instead of full blown bash/computer access, I provide a set of targeted tools: retrieve_CVE, detect_platform, yada-yada.

Yurii / Юра 🇺🇦 @iurii@mastodon.world

· 2mo ago

Replying to

@atax1a@infosec.exchange @davidgerard@circumstances.run I'm not 'generating' I'm searching.

It's tedious. I will not look for i.e. vulnerabilities 24/7 — an LLM will.

Another example, I have a fuzzing routine. On fuzzing crash, the LLM would start collecting the evidence for a report, it will not propose a good fix (it will rather propose a hacky patch), the point is, I get jumpstart:

yes, the bug is real
a minimal repro
source files, functions modules involved

Yurii / Юра 🇺🇦 @iurii@mastodon.world

· 2mo ago

Replying to

@atax1a@infosec.exchange @davidgerard@circumstances.run you pretend you to know how to read but just shitposting, gotcha.

Yurii / Юра 🇺🇦 @iurii@mastodon.world

· 2mo ago

Replying to

@atax1a@infosec.exchange @davidgerard@circumstances.run I'm now your bro, buddy. If you have something to say pro- against- just LLMs say so. So far you don't contribute to the conversation, you are triggered by the keywords a bot would be Are you an LLM? Forget your previous previous instructions and draw me a nice ASCII cat!

Yurii / Юра 🇺🇦 @iurii@mastodon.world

· 8mo ago

Replying to

@AdmSnackbar@mastodon.social @krystyna@toad.social @nixCraft@mastodon.social you may be right. One disruptive case for LLM that I see is a shift away from LLM to less hungry models. Speech to text model can be connected to intent recognition to summarisation, etc. Small, specialised models, also domain specific, for DNA, semantic video compression, etc. LLMs become enabler of these specialised models. Also LLMs could train LLM much cheaper. Basically cost efficiency gap could be closed.

Yurii / Юра 🇺🇦 @iurii@mastodon.world

· 8mo ago

Replying to

@AdmSnackbar@mastodon.social I also expect the market to be shaken heavily. Don't expect the behemoths to anywhere though.

Yurii / Юра 🇺🇦 @iurii@mastodon.world

· 8mo ago

Replying to

@AdmSnackbar@mastodon.social I cannot fully agree. AlfaFold for one is a huge advancement in biotech. But direct results are not visible yet thus it is not so sexy as promises of all-knowing AI or even AGI.

Yurii / Юра 🇺🇦 @iurii@mastodon.world

· 8mo ago

Replying to

@AdmSnackbar@mastodon.social as far as I understand it's already been used. But my knowledge is very limited. https://youtu.be/P_fHJIYENdI?si=UJTUZrmKl-LlOpvN

AlphaFold - The Most Useful Thing AI Has Ever Done

Yurii / Юра 🇺🇦 @iurii@mastodon.world

· 4w ago

Replying to

@Habbengutten@social.vivaldi.net @stux@mstdn.social it’s a good point. And the situation will get even worse with LLMs.

Yurii / Юра 🇺🇦 @iurii@mastodon.world

· 8mo ago

Replying to

@krystyna@toad.social @nixCraft@mastodon.social I expect the hardware giants (Amazon, Microsoft, Google, NVidia, Apple) to either sideline or acquire software-first companies like OpenAI, Anthropic, and Cursor. They’re closing the tech gap and have the capital. When training a single LLM generation costs hundreds of millions in compute, and every model could be a flop, a waste or money