devsimsek

Hardware and software enthusiast.

Pixelfed: Devsimsek (npub1kqh…ge6m)

Blog: https://smsk.dev
Github: https://github.com/devsimsek

Public Key

npub17gfm6zn95clryx3jrsnjvr4dernuuj6405enven3xxwfw2l7vvks7ua879

NIP-05 Address

[email protected]

Profile Code

nprofile1qqs0yyaapfj6v03jrgepcfexp6ku3e7wfd2h6vekvecnr8yh90lxxtgpzpmhxue69uhkummnw3ezuamfdejszrthwden5te0dehhxtnvdakq32dlwz

Publishing to

Author Public Key

npub17gfm6zn95clryx3jrsnjvr4dernuuj6405enven3xxwfw2l7vvks7ua879

Show more details

Last Notes

2026-04-26 21:01:57 CEST

Researchers just mathematically proved that AI can't recursively self-improve its way to superintelligence.
Not "we think it's unlikely." Not "it seems hard." Formally proved.
The model doesn't climb toward AGI — it slowly forgets what reality looks like. They call it model collapse. The math calls it inevitable.
I wrote about it 👇
https://smsk.dev/2026/04/26/ai-cannot-self-improve-and-math-behind-proves-it/
#AI #MachineLearning #LLM #Research

2026-04-19 21:25:53 CEST

- reply

that list doesn't look so good...

2026-04-15 11:58:42 CEST

watched a youtube video about cleaning camera sensors.
obviously the correct response was to spend a weekend building a PWA.
introducing Speck — sensor dust visualization, mapping, and EXIF reading. offline-capable, no install needed.
https://github.com/devsimsek/Speck
https://devsimsek.github.io/Speck
#WebDev #Photography #PWA #OpenSource
https://media.universeodon.com/media_attachments/files/116/408/123/651/854/872/original/787b36a243621c12.png

2026-04-15 11:58:42 CEST

- reply

the trick to finding sensor dust: grayscale + invert + crank contrast to 0–2000%.
suddenly every speck you've been ignoring for months becomes painfully obvious.
built that filter into a canvas-based PWA so you never have to do it manually in Lightroom again.
#CanvasAPI #JavaScript #Photography

2026-04-14 01:10:18 CEST

- reply

Now, I just can't see it. I miss my earlier days... (where I had my innocence :D)

2026-04-13 17:33:11 CEST

- reply

5/5 What's your self-hosted photo setup? Instagram refugees assemble.
Would you use a TUI CLI for albums?
Source: github.com/devsimsek/TypeGrid
Oh, also I've written another article about this check it out: https://smsk.dev/2026/04/13/typegrid-i-built-my-own-photo-portfolio-because-instagram-sucks-and-i-was-bored/

2026-04-13 17:33:11 CEST

- reply

4/5 Why? My love of capturing moments and too much free time.
Check it here: https://kare.smsk.dev
#webdev #sideproject #alpinejs #TypeGrid

2026-04-13 17:33:11 CEST

- reply

2/5 The CLI screenshot says it all:
📁 Albums | 🖼️ Images | 📊 EXIF & metadata
Rosé Pine colors, Zed-inspired, keyboard-first (hjkl).
Generates WebP thumbs, XML sitemaps, OpenGraph. Even GC.
https://github.com/devsimsek/TypeGrid
https://media.universeodon.com/media_attachments/files/116/398/067/316/299/136/original/7e221af17496db23.png

2026-04-13 17:33:11 CEST

- reply

3/5 Architecture:
Photos → TypeGrid CLI → JSON API → Alpine SPA → static host
No build step. Extend with your own dashboard if you want.
CLI is one implementation of the TypeGrid spec — open source, build your own.

2026-04-13 17:33:10 CEST

1/5 I was bored, hated Instagram, found Pixelfed slow → built my own static photo portfolio.
Meet **TypeGrid**: JSON headless API + Alpine.js frontend + TUI CLI that does *everything*.
https://kare.smsk.dev #photography #opensource #cli
https://media.universeodon.com/media_attachments/files/116/398/121/563/248/748/original/9e9019231036d581.png

2026-04-02 17:12:14 CEST

- reply

What? I am flabbergasted...

2026-04-02 00:41:59 CEST

just dropped a new post 🍵
went full nerd mode on self-hosting git — starting from bare repos & SSH, through gitolite, all the way to a full gitea install on alpine linux 🏔️
also roasted gitlab (lovingly), compared gitea vs forgejo vs gogs vs the whole gang
spoiler: gitea won. it's written in go. you already know.
📖 https://smsk.dev/2026/04/02/self-hosting-git-from-bare-repos-to-gitea-on-alpine/
#git #selfhosted #gitea #alpinelinux #devops #opensource #golang
https://media.universeodon.com/media_attachments/files/116/331/860/172/000/947/original/f845e6d1f720632d.jpeg

2026-03-30 22:28:57 CEST

- reply

maybe I can do that as well, well we will see...

2026-03-26 23:36:02 CET

- reply

The attacker compromised Trivy — LiteLLM's own
CI/CD security scanner — and used it to push
malicious packages to PyPI under a maintainer's account.
v1.82.8 even dropped a persistence file into
site-packages that survives uninstall.
The security tool. Was the attack vector.

2026-03-26 23:36:02 CET

- reply

Hot take: this isn't an "AI" problem.
ML and deep learning are math. Gradient descent
doesn't steal your secrets.
What steals your secrets is a rushed,
overhyped ecosystem where devs blindly
`pip install` without reading a changelog.
I review every dependency update before I trust it.
It's boring. It works.

2026-03-26 23:36:02 CET

- reply

I wrote up the full breakdown — what happened,
what was stolen, the uncomfortable pattern behind
incidents like this, and what to do if you're affected.
👉 https://smsk.dev/2026/03/26/the-litellm-breach-why-i-dont-integrate-ai-into-my-apps-and-you-might-want-to-reconsider-too/
If your stack depends on LLM middleware,
this one's worth a read. 🔒

2026-03-26 23:36:01 CET

A supply chain attack hit LiteLLM on March 24th.
Two PyPI versions. Six hours. Your AWS keys, SSH keys,
DB passwords, Kubernetes tokens — silently exfiltrated.
You didn't even have to install it directly.
A transitive dependency was enough.
EVERYTHING FROM A TRANSITIVE DEPENDENCY!

2026-03-21 18:02:48 CET

- reply

looks like a picture i've taken to be honest :)
https://media.universeodon.com/media_attachments/files/116/268/240/172/160/889/original/d92b2a37d232d87a.jpg

2026-03-20 17:09:15 CET

- reply

the funny thing is; ai is slowly killing google search. They can't gain revenue from AdSense like before.

2026-03-20 16:52:26 CET

Usually, I catch up on tech and science via newsletters during my commute, but today's deep dive into Amazon’s current state has me reeling.
I really respected the AWS CEO’s previous stance that juniors are essential and replacing them with AI is a mistake. But looking at the "solutions" being pushed now? It feels like the wheels are coming off.
Senior devs are being turned into human filters for an endless stream of AI-generated code and PRs.
It’s creating massive "cognitive debt" where no one actually knows how the systems work anymore. Between accidental environment deletions (they deserved it tbh) and massive outages, I have to ask: Is FAANG (minus Apple, maybe because of neo) even sane anymore?
PS: We as humanity need to learn one thing, you can't truly replace a person. Stop replacing people with AI!

2026-03-19 16:17:17 CET

- reply

hmmmmm we can use this modal to make hungry hungry hippos :)))

2026-03-18 10:52:44 CET

- reply

Looks awesome tbf :D

2026-03-15 23:27:57 CET

- reply

"We Can't Have Nice Things Anymore" - 🧵 2/4
F1 2026 regs were supposed to be revolutionary. Instead, the fastest man on the grid is emotionally detaching from the sport and warning the rules "aren't ready for prime time."
We took a technological marvel — a car that corners at 5G, a human pushed to physical limits — and buried it under committee decisions and political compromise.
This is a metaphor. A very fast, very sad metaphor.

2026-03-15 23:27:57 CET

"We Can't Have Nice Things Anymore" - 🧵 1/4
We used to build things that made jaws drop. The Concorde. The Space Shuttle. The F1 cars of the turbo era. Machines so insane the rulebooks had to *ban* them.
Now Max Verstappen (a four-time world champion, and my favourite driver) is complaining his 2026 car "has no balance" and called his sprint race weekend "a disaster."
We just can't have nice things anymore.
https://media.universeodon.com/media_attachments/files/116/235/596/394/391/357/original/2dffb9c8dbaac043.jpg

2026-03-15 23:27:57 CET

- reply

"We Can't Have Nice Things Anymore" - 🧵 4/4
Pattern recognition: every time we build something extraordinary, we regulate it to mediocrity, outsource it to automation, or drown it in committee consensus.
F1 banned the most brilliant cars. AI is replacing the most brilliant thinking.
Creative thinking tests show the average person now generates 3–5 unique ideas vs. 8–12 before widespread AI adoption.
We are becoming spectators of our own potential.
Fix your rulesets. Use your brain. Read more. 🏎️🧠📚 (Please, this may be our last call)

2026-03-15 23:27:57 CET

- reply

"We Can't Have Nice Things Anymore" - 🧵 3/4
**And it's not just F1.**
MIT ran a study. Students who used ChatGPT showed "reduced memory retention, diminished brain activity, and performed worse at all levels: neural, linguistic, and scoring." I can also confirm this finding as a mentor to many youngsters.
We built the most powerful cognitive tool in human history — and we're using it to *stop thinking.*
A University of Toronto study found that even after people *quit* using AI, their lowered creativity persisted.
Cognitive atrophy. In real time.
---
Ref.
https://www.media.mit.edu/publications/your-brain-on-chatgpt/
https://www.unleash.ai/artificial-intelligence/is-ai-causing-a-decline-in-cognitive-and-creative-skills/

2026-03-15 23:19:46 CET

- reply

I think I like this one more than the others.

2025-12-11 14:22:08 CET

- reply

Whoa, that's so cool :) His work really did influence people.

2025-12-10 22:32:39 CET

- reply

True, he's immortal.