So, talking about people on Reddit not having humor; i got banned by my bank (bunq) forever. I complained about my always unasked for planted tree after every expensive I do of x euro's. I've planted 33 or something. So, at least I posted that I want to have a picture of my tree. Many others wanted pictures of their trees too and some actually had whole woods of trees actually :P

So, nobody got a picture of their tree and I got banned.

But now I think about it, if I alone are already responsible for 33 trees and some people own complete woods appearantly. Is it actually possible that they're really planting all those trees? They're a bank, they're massive, it must be the biggest tree planting operation ever. Gonna research this.

SInce this app has no selection regarding rant type, let's see what it will become.

So I saw a new fucked up trick a job offering company did. If you view their job offers and want to click on one of them it forces you to sign up to their newsletter. No opt out in small lettering.

What kind of fucked up place forces opting into a news letter to view a job offer?

Job searching in 2018. Not great options, but job providers and sites are still human.

Job searching in 2025. Process is completely devoid of human decency, filled with cunts, and makes you want to die early. Plus there are NO FUCKING JOBS.

For the poor devRant android app users that can't visit devRant anymore (like myself): https://rantii.molodetz.nl

It's a simple devRant frontend and it has a few features:
- markdown (# Titles and ``` for code ssupported!)
- syntax highlighting
- youtube / image embedding / url preview
- install (in that case it's a nice full screen app), you can find this under settings.
- click on notification and it will bring you to the exact correct message
- missing rant type select box. Exclusive!

Like always, open source: https://retoor.molodetz.nl/retoor/...

It's mainly for myself actually but let me know if you use it as well, then I will take that into account when making changes / ensure uptime.

@lensflare I'm doing AOC with swift so far I like it, it's unfortunate they do the retarded "modern" `let name: <type>` thing and strings can be annoying to substring, otherwise it's good

It's so funny how customers like the small and unimportant things. They are impressed by a display of skill rather than skill.

For instance, I needed to quickly create a password. So, I opened my terminal, typed in as always

dd if=/dev/urandom count=500 | base64 -w 0

Then I copied a random chunk of some 200 to 500 chars and copied it as the temporary password and sent it around.

I was told later on that I impressed with my technical know-how and skills...

But I guess, now that I impressed, I can profit of the halo effect.

Guess, take this as PSA. Sound confident. That's more important than being confident. They have no idea what's going on.

Tell me the truth.

Do the curtains match the carpet?

I am so assertive, that if you find yourself intimidated by everything i say, you're a pussy. Just a fact. Anyway, nice day for fishing.

Talking about fishes, is there some @blindxfish in them house? I really love my.devplace.net.

It's really great, i hope that we all will share our development experiences there one day.

But still, for chat there's still nothing available like Snek. When you go Snek, you never go Beck!
'
Hihi.

I'm a bit drunk and the rest. I prefer the rest and then to rest.

Dobranich.

My manager uses Claude chatbot for writing all our user stories while we're on the call with him. It spits out so much shit! He'll say, "let's make sure it hasn't hallucinated anything .. nope"

There's no way he can be reading all that. I'll call out stuff early on, and he'll say "that's wrong, it hallucinated," and fix stuff and remove half of it, but 45 minutes into the story session, I think everyone tunes out.

Implementing this is going to be fucking hell.

Made this little example to show the difference in codegen quality using Gc smart pointers versus reference counted Rc's can make

Actually pretty crazy:

https://godbolt.org/z/vsTeGcz8M

Just read somebody say "she climbs like a homesick angel" to a fighter jet flying straight up and I think that's one of the most beautiful ways to describe it

hello thx for applying to our startup :) we are looking to hire an engineer asap so let's get things in motion. We will only interview candidates that pass our custom take home challenge, which is designed to test your ability to solve real problems.

Anyone who passes our take home challenge immediately gets an interview with our CTO and thus has a high likelihood of being hired.

Vibe coding warning:
If you cannot write code without AI assistance, you will not pass our live technical interview so please do not waste your time now attempting this challenge; we won't hire you.

Here is the take home challenge:
Figure out how to programmatically send disappearing photos on Telegram web using a chrome extension.

What we measure with this take home challenge:
This challenge is designed to measure your resourcefulness and ability to quickly learn new things. Being an expert on chrome extensions or javascript won't help much.

More details:
Telegram on mobile allows you to send disappearing photos, but Telegram Web doesn't expose sending disappearing photos in the UI. You'll have to dig deeper.

Rules (!important):
Your submission will not count if it breaks any of these rules.
1) Your solution must be fully contained in a chrome extension. This means no other code (ie. localhost server) outside of the extension is allowed.
2) You are not allowed to bundle outside libraries (ie. gramjs, mtproto, mqtt, etc)
3) You must be on web.telegram .org/a which is version A of Telegram web, not web.telegram .org/k which is version K or others.
4) Your solution must be able to send photos programmatically and not require manually sending photos in the UI

Expectations:
Although a solution that just follows the rules will get you an interview, the quality of your submission will be a key factor in our hiring decision. We expect your code to be clean, production ready, and easy to review. This means removing all unused code and files you may have added during experimentation/testing.

Video guide:
Don't let the high view count on the video scare you, if you are seeing this we are still accepting submissions.
Please watch the video I made describing the challenge and showcasing the solution

Here is the job posting in case you lost it:

Here is the code for the challenge:
You are given a simple boilerplate chrome extension, feel free to change any file or add new files as needed

Submission:
Do NOT upload your submission to GitHub, you wouldn't want others stealing your work and submitting it. Please reply to this only with a public **Google Drive** link containing:
-Your chrome extension code as a **folder** (not as a zip or rar file), this makes it easier for us to review
-A short screen recording showing your solution works

Deadline:
Please try to finish by the end of day Friday (December 5); speed is one of the things we are measuring here too. We will not make hiring decisions before this date so please attempt the challenge if you are seeing this before the deadline.

Communication:
Please only reply to this message with your Google Drive link submission, otherwise I will ignore and assume you are an AI. If you have questions, please figure them out on your own as that is part of the challenge.

Good luck, and I hope to hear from you soon :)

For people who need a https version of dR, it's here: https://dr.molodetz.nl.
Does not monitor, source code of proxy here: https://molodetz.nl/project/rproxy/....
It only counts amount of data.
Graphs are cool: https://dr.molodetz.nl/rproxy/...

It's actually a fun proxy to use in docker setups. The configuration is very easy (json file as described on page). But it does not forward based on path or something else. Only on host name. Maybe i'll implement it in the future for the sake of completeness.

The dashboard/reverse proxy is written because my friend had such nice dashboard and he said "Yeah, it's cool but heavy as fuck". And i was like: that does not make sense. So made a feather light one. Could be a bit lighter, i have to tweak sqlite a bit. After a few hundred gb, the server starts to boot slower, i store too much data. Does not affect operation tho.

Also has rate limiting (not much used in reality yet). Not sure how good it is.

Already quite some traffic goes trough it.

Sid, like always, fuck you and I hope your elephant dies. Idiot. Guy with PMS.

Ah fuck my asshole of course it wasn't that easy

Great that my GC can detect any managed pointer off the stack et al. Yeah but what if a GC ptr is behind some random non-gc allocation, inside a future, etc.

fuckkkk

Best idea right now I replacing the global allocator. Then I know all allocations and can conservatively still traverse them (means way more checks... ugh). And if one allloc is a gc object then I know its vtable and can do precise analysis

Really not perfect. But guess what: boehm does the same lol

I should really stop being that dismissive of myself

I had yesterday a meeting with the backend-dev/manager. He at least mentioned that GraphQL was not the way to go and that he has no idea how to tackle certain requirements. He is planning to sit together with the internal future users to check what data is truely needed (fyi, he started the project 11 months ago).

So yeah, the project is now on hold and he assigned me a 2h task for the rest of this week; fun.

I've been thinking a lot about garbage collectors for compiled languages again

Funnily enough my idea for a multi threaded gc for Rust is basically how boehm does it too (well at least on linux): On a GC cycle you send a unix signal to all other threads to make them pause and then in the signal handler you save all regs, then walk the stack and regs upwards over 8 byte values. Any value that falls into the gc heap and is a valid allocation is assumed to be a ptr and that object gets marked

Pretty clever

Though I'm working on a single threaded GC for Rust now. That one actually has a real use case too: I don't have to use Rc<T> and .clone() them all over the place in my compiler! Just beautiful wrapped pointers with copy semantics 😌

Did you know that if you boil a funny bone, it becomes a laughing stock?

Week: 116 (Year 3)

Plans for the weekend?

Question: How are you gonna enjoy the upcoming vacation? Assuming December is a chilling month for most of you

last Weekend : https://devrant.com/rants/19386937

Oh no, devrant cert just expired...

If we don't build AGI, the .govs and corpos of the world will build it first, and then use it against us.

Just saying

Should an RPG exist that normalizes skills somehow? You can increase skills, but it doesn't make some enemies so easy they just fall down after one hit. I get how in D&D that you can tank more the higher level. But it was rare to increase attributes. Now it seems every RPG increases attributes. So early enemies fall down very easy.

How to keep things challenging without making enemies too easy? At some point everything is one hit and dead.

I wonder if technical knowledge truly matters outside of those companies that try to filter people based on tests...

And if not... then I'm just studying for myself. lol

Code tells a story.

It is usually not especially interesting. It is not captivating. But it is a story.

The story often goes like this: Instead of using the class directly, an interface was used. Someone expects this implementation to change a lot. Or they want to break the logic of the code apart. Speed up compilation times. Maybe they plan on having multiple different implementations.

That's, well... no, it is not exciting. But sometimes you look at it and wonder why. And you muse to yourself, what could be the reason for this curious choice. Is the whole project written like that? Maybe two developer developed the feature together and they agreed at the very beginning to write the interface in order to be able to cleanly work in parallel. I did this with a colleague. We worked on a common feature. We decided to write an interface first, then we worked separately and in parallel on the two sides of the interface. In the end we left it in and we were asked why we decided to use an interface.

Code tells a story. But so very often now, the story I hear when I ask why is "copilot did it." Fuck you. And fuck your "hey, copilot says you could do something better" in merge reviews.

I don't care if you want to use AI. I don't think it is good. You dull your skills. But hey, I don't really care about it. I don't care if you decide to use a plastic fork to remove your left eye. I mean, I would have recommended a sharper instrument, but fuck, if you want to mutilate yourself. Go for it. But when git blame tells me it was you who wrote the code, then don't fucking excuse any decision with "copilot did it." And when copilot tells you there is an improvement in my merge request, you don't need to tell me that it came from copilot. That is no authority. If you think copilot is right, you argue for it, and you give me arguments and reasons. But if you don't even understand yourself why it is supposedly better, just fuck off.

guess who's back?!

soooo... long story short... i got super ill... like so ill i dont wanna give many details cuz
1. rather not bum you entertaining weirdos out... that'd make this bittersweet.
2. i dont wanna get a bunch of emo comments of best wishes etc

it's fixed, enough, now.

just to be clear, "ill" is not some paraphrasing for mentally ill/unstable/depressed/whatever else may pop into your heads... nor is it code for something like substance abuse/rehab or incarceration, etc... there was definitely entropy at play so it doesnt need more =P

So I started typing "// putain de bor..." in a comment because it's one of those sort of mornings.

The autocomplete suggested "putain de bordel de merde ce que c'est con javascript a la con".

I don't know much French but on the face of it that sounds like a reasonable assessment.

Besides keeping your computer unlocked for long-running tasks, Caffeine can also be used to fool MS Teams' activity monitor.

Fuck, I had an initial screening on Tuesday. Recruiter said they would send me an email so I can look over some stuff this weekend. No email. I tried messaging them on Indeed where they contacted me about sending them a targeted CV. Nothing. I found them on linkedin and requested a connect. Nothing. I am trying to be relaxed about this, but I will be pissed if they are ghosting me. They need someone before January. Bitch, I am right here!

Currently working on:

Conversion of existing models to full closed form fusion representations. Built a toy example with just the dense layers, got it up to 65m parameters with minimal accuracy loss and almost 80% reduction in memory requirements.

Figured out how to also handle convolution layers in this new representation, and am figuring out skip connections this week so I can implement an example of the full closed form variety. All the math checks out.

Full CFF should let me reduce memory requirements anywhere from ten times to fifty times whats typical, as well as similar improvements in inference times and compute.

The partial implementation also still allows for training.

I'm working toward a demonstration of GPT-2 training and running on a consumer desktop, without quantization.

Whether it will work out, who knows¯\_(ツ)_/¯.

But if it does, we're greenlight on the research lab being launched in late 2026. And when it does, I'll have to steal a couple of you as 'entourage'/VIP hires because devranters are really the only group of people I like to hang out and do things with anymore.

Oh dang you guys have been busy enough that there are new rants all the way at the end of the recents list

Another small win and a first world problem:

I got another onsite for another company, but now I need to figure out how to line these interviews up. I am doing one next week on Thursday, and trying to do the new onsite the coming Monday.

Is that too far apart? I was tempted to have it on Friday, but two days of interviews back to back seems super tiring. I also need to brush up a bit more on system design (I am not amazing at it coming from a mostly frontend background).

We need a definition for when something is AI slop or not because people just throwing the word around when something is AI generated while there still is often much work behind it. And with much work, i do not mean just entering thousand prompts to get something right. That's actually AI abuse imho.

So, proposals, when is something AI slop? What is the definition?