Keeping up the tradition!
https://devrant.com/rants/15030806/...

Now powered by the awesome Claude 4!!!! The latest bleeding edge gem in the LLM trend that by the bold claims of big tech and various youtubers is stunning and will replace programmers/insane/godsmacking

This time, I decided to post because I was on the verge of a mental breakdown and I had a firsthand experience and a bit of free time, so instead of waiting that the rage boils down, I just took the chance and so we have a fresh AI experience to proof my previous rants.

Problem:
I have an application that manages interactions between a Mediator pattern between Kafka, some http listeners and other stuff (Rabbitmq, Redis (redis= another shit) etc etc)

Kafka handling was already done, i needed to exacly replicate what was being done with kafka for http. I said, heck, we have AI, it's good to do bovine stupid work, why not let him copy the whole structure, they have also agents on copilot on vs code, and they pulled out Claude 4, let's give it faith again.

Big error.
Beside the havoc the fucking agentic AI not being able to fully copy an already existing pattern and scaffolding, it did half shit wrong.

One of the most frustrating parts, since today I have an example, was this (summarized version of the prompt because I was very much more specific also giving the AI files and pointers to the code and explaining it):

PROMPT:"I have to return fake http transports, just copy what exists for kafka and adapt it to a new transport type called "Http", this is the structure {json with the class}

AI: "I need to see the current ITransportsItemResponse interface first to understand the structure, then I'll modify ...... BLAH BLAH"

Of course I had to review the whole code.

PROMPT:"any valid reason why you dumb shithead didn't respect the IGetKafkaConnectionsItemResponse and generic IGetTransportsItemResponse format and you pulled out of your ass a full blown IGetHttpConnectionsItemResponse with extra properties when compared to IGetKafkaConnectionsItemResponse"

AI: "You're absolutely right, I fucked up. Looking at the IGetHttpConnectionsItemResponse interface, it has way more properties than the basic IGetTransportsItemResponse and IGetKafkaConnectionsItemResponse interfaces. BLAH BLAH wall of text reasoning"

Then it modified the api calls to suit the kafka objects and didn't update any class of the Http transports

PROMPT:"So you tought it was a good idea to leave the #sym:IGetHttpConnectionsItemResponse like it is respect to #sym:IGetKafkaConnectionsItemResponse ? Why? And answer the why don't apologize like a little wuss"

AI: "I left them different because they represent different domain concepts... BLAH BLAH EXPLANATIONS WALL OF TEXT"

Now ai tought it was good to do this and that but....

PROMPT: "In your fucking careful analysis didn't you reallize the "items" object are a basic representation and in the fuckin same folder there is also the object with the full set of properties?"

Literally it missed objects used in the same folder for different scopes and modified stuff without caring.

AI: "You're absolutely right. I missed that there are TWO different response types BLA BLAH"

I won't continue to not get too lenghty than it already is but the point is:

AI IS RETARDED.
People say it will replace programmers.
People says agents are the future.

Sad reality it's an overglorified broken ball of if/else that can't do shit well beside bovine work.

No amount of tutoring it with careful prompts, explainig the code and whatever else is going to fix it.
I've used gpt since gpt 3 and no model has been up to anything good, not even NLP. They suck also at the sole scope they were invented for.

I tried to ask GPT to make a curriculum based on another, I gave it the example curriculum and another one with the informations.

I carefully explained that it must not be a copy of the other, they are 2 different roles and to play by fantasy to make it look it was written by 2 different persons and to not copy stuff from the other.

Hope lost. It looked like the other curriculum was copied over and some words swapped, lol.

What a fucking joke, lmao, I am studying deep learning and machine learning to get on the bandwagon to make my professional figure more appealing, but I can already feel this is a waste of time.

Today I almost had an engineering orgasm. I'm designing a Eurocard PCB for a Subrack. All components already came with a 3D model within the EDA software. I could therefor export the entire PCB to an STL and print it out for a fitment check. Also the Subrack Vendor has an online designer for the front. You guessed it! It can be exported to CAD and Printed. Man I LOVE the current state of electrical and mechanical engineering. This is such a fucking game changer! I really hope the electrical side checks out thou as prototype manufacturing is expensive! Like 4k can go down the drain for just 5 boards.

Finally got all the php errors fixed on my servers. Feels good man.

Enter BIOS menu
See the option called something like “ionic capacitor discombobulation”
Click “help” near the option to learn what the hell that is
“Enables or disables ionic capacitor discombobulation”
ffs

I hate what AI has done to developers, man.

I was discussing something a couple of days ago while my colleague was sharing his screen. He (Android/Flutter developer asked for my help for something custom in the iOS build) And while we were discussing it, he went to ChatGPT.

He wrote a bad prompt using wrong terms (for example how tabbar in ios is a different thing than android), i told him he was wrong and what the solution is, he didn't listen and went to try out what the AI said, made a bunch of errors, the proceeded to copy the errors to chatgpt wasting so much time.

AI to some developers is like tiktok to kids.

Stack Overflow has the best April fools prank!

They are doing an 8+ hour database migration on a business day during business hours!

The prank is the fact that it’s not a joke! They really decided this for some reason!

Cars 1 was NOT released 19 years ago. No, it just wasn't. Fuck you, it wasnt. I dont care about your facts. It's all fake. You're litearlly just lying. No it cannot have come out 19 years ago. It just couldn't have, thats way to long ago. You're just weird. Calenders are just wrong

Adaptive Latent Hypersurfaces

The idea is rather than adjusting embedding latents, we learn a model that takes
the context tokens as input, and generates an efficient adapter or transform of the latents,
so when the latents are grabbed for that same input, they produce outputs with much lower perplexity and loss.

This can be trained autoregressively.
This is similar in some respects to hypernetworks, but applied to embeddings.

The thinking is we shouldn't change latents directly, because any given vector will general be orthogonal to any other, and changing the latents introduces variance for some subset of other inputs over some distribution that is partially or fully out-of-distribution to the current training and verification data sets, thus ultimately leading to a plateau in loss-drop.

Therefore, by autoregressively taking an input, and learning a model that produces a transform on the latents of a token dictionary, we can avoid this ossification of global minima, by finding hypersurfaces that adapt the embeddings, rather than changing them directly.

The result is a network that essentially acts a a compressor of all relevant use cases, without leading to overfitting on in-distribution data and underfitting on out-of-distribution data.

s3 storage sucks. Requires you to implement special stuff in software. The s3fs is a terrible thing not supporting many native file operations. For example, i can't cp * /to-target. It can't do nothing.

So, that was a short adventure. Now i found: https://interserver.net/r/1035270/ (affiliate link, why not). 3,- 1tb storage WITH complete vps. 1gb / 1 core but it does great as sshfs drive. It works so fast, i gonna host complete projects on it like https://devrant.molodetz.nl. sshfs drives really have the comfort of a native drive. Boarded just uploaded 600Mb to snek in two minutes. That movie was completely streamable in the snek web interface, you could comfortably click trough the video.

So, time to make a custom script to push all data from s3bucket (blegh!) to this beautiful drive.

No amount of sub-par developers cheering for AI to overtake them will be enough to make me install Copilot or any glorified autocomplete. None of that would be needed anyway if people documented and tested their shit anyway.

My first prompt with grok, for giggles:
"how to disable grok"
This is a certified black mirror classic.

moving from Linode to Contabo because my shitty personal projects do not need expensive uptime guarantees.

Why would you offer DNS in the year of our Lord 2025 without any kind of API?

I always felt a bit bad for scammers. That your life must suck so hard that you screw over people. But now i watched them working on YouTube and i don't feel bad for them at all anymore. First of all, trash people, the lowest of the lowest. Second - they show no mercy to their victims. And that shouting and stuff. Is that normal in India? If you do that to a Dutch, it won't engage a phone call with you at all. They start the phone calls in a fucked up way already.

Trash people.

@dfox @trogus please update the native android apk on playstore to android 15/16 . it isn't showing in playstore anymore and downloading from unsafe sources feels like a security risk

Does anyone else find the raspberry pi / Linux boot console output comforting?

Don't want a boot loading bar Mac, I want console output

The more AI claims that GIMP has no connection to either a Gimp eg person with a limp or the Gimp from pulp fiction the more knowing that age group and the fact that pulp fiction came out in 1994 and the project release in 1996 and was written by two COLLEGE STUDENTS the more convinced I am that they made ai intentionally stupid and that it needs to be removed from the hands that prevent it from being used to describe things accurately

My server is under heavy attack, many IP's.

Here's a part of requests per ip all in same range:
```
509: 47.82.11.166
509: 193.41.206.202
512: 47.82.11.183
521: 47.82.10.30
548: 47.82.11.8
553: 47.82.11.177
559: 47.82.11.230
606: 47.82.11.41
659: 47.82.11.249
675: 47.82.10.139
684: 136.243.228.177
700: 47.82.11.27
865: 85.25.210.70
```

Al together, but my server died for a second. I thought, how is that possible? All my request limiters gone somehow. I have no idea how that's possible. Implemented a few days ago and tested it.

Meh, server is handling quite well, i have time to write this rant before to fix it. But dammit. I tested it together with people on Snek. I'm flabbergasted.

Fucking scriptkiddies. YOu were lucky.. Somehow..

Will disable ping, that's where the issues start with i guess.

Devrant servers are still running? Oh man it's been a while

Past hour five spam messages came in and last 24h around 20. Many of them seem unique, I'm so interested, how is this even possible. Many of them even have a phone number attached to it. All that domains, emailadresses and phone numbers. Omg. Why all these small brands instead of a big one when all sites kinda have the same formula? It's such a mystery.

Looks like image uploading is dead again. :(

Terry Davis used algorithms to worship god. Now we worship the algorithm.

I'm a proud man, but I'm not such a fan of my own farts that I can't admit when I maybe should have listened to other people's advice earlier.

After having been convinced of the scope of impact of AI by someone I respect, I started playing with the Jetbrains AI tools. I am impressed at its ability to process my code and give actually helpful input and to consolidate documentation into a form that is concise and helpful.

I finally get what people mean when they say it saves time.

A couple things that truly warmed me up to it is, one:
I wanted to know if I could return a string, float array, whatever, from c++ to a python script. I was assured the answer was yes, but I just COULD NOT get it to work, so I gave up on it. I asked the question to the Jetbrains AI (4o in this case) and it gave me what I needed, and now I can return a string from c++ to python no problem. There are a lot of little questions like this that I gave up on that I now have to explore again, which is both exciting and annoying, because I already have a thousand hobbies.

And two:
I am working on an html email. It's a mess of tables and text and inline styles. Compared to the markup I'm used to writing, it's tedious to trudge through to make even simple changes. I was able to successfully instruct it to make a specific copy change, while respecting the document's indenting, all on the first try.

I will forever maintain that it will enable a generation of drones that don't understand how to do simple things and will atrophy the skills of otherwise capable people that use it as a crutch.

I will also always maintain that its foundation is built on mass theft and is a monument to the uneven application of intellectual property protection laws. But with DeepSeek coming out and having done the same thing to them, I find myself enjoying the turnabout. I'm also amused that I coincidentally jumped into the pool right as things got interesting.

All that said, as a reference tool, like Google and Wikipedia used to be, it's not the force of pure evil I held it as. It is actually very useful and if used responsibly on an individual level, can be an amazing productivity tool and can even teach its users new things.

A 5-minute tutorial to register a fucking timesheet in your old-ass corporate shit tool: https://vimeo.com/1045331789/.... And I kid you not you need to convert worked hours to days so if you worked 6 hours write 0.75. Also it needs to be submitted 5-10 days in advance of the end of the month. If you are 1 hour late they will spam you and your manager, like WTF

This corporate behemoth of an IT consulting company I do (sub)contracting via is really stretching how draconian its timesheet policies can be. This is only one of the 4 timesheets I need to manually sync every month and coincidentally care the least about.

Every year they keep adding extra dumb rules that supposedly help managing their records. The moral of the story is I might quit my job for the first time due to the intermediate company's timesheets policies.

Week : 72 ( Year 1 )

How was the weekend?

What’s something a guy/girl has done (intentional or not) that instantly made you think, “Wow, he/she is different in a good way?

Previous Week : https://devrant.com/rants/12627389

I wonder whether this has gotten here yet but it knocked me out. I don't write in trap English but I'm a sucker for memes. Imagine how bored he must have been to start documenting legacy code without the urge to refactor. Just "vibing"

I opened the backlog today and found a bug assigned to me.

the bug basically describing that the images are not loading on the app, I take one look at the response and find the images url are sent as ip address.

I mention this to the tester and with a straight face they say "well why is it working on my laptop?" the laptop that's connected to their company's network.

god why am I even doing this.

Here is my idea for a time machine which can only send one bit of information back in time.

@Wisecrack has asked me about it and I didn’t want to write it in comments because of the character limit.

So here we go.

The DCQE (delayed-choice quantum eraser) is an experiment that has been successfully performed by many people in small scale.
You can read about it on wikipedia but I'll try to explain it here.
https://en.wikipedia.org/wiki/...

First I need to quickly explain the double slit experiment because DCQE is based on that.
The double slit experiment shows that a particle, like a photon, seems to go through both slits at the same time and interfere with itself as a wave to finally contribute to an interference pattern when hit on a screen. Many photons will result in a visible interference pattern.
However, if we install a detector somewhere between the particle emitter an the screen, so that we know which path the particle must have taken (which slit it has passed through), then there will be no interference pattern on the screen because the particle will not behave as a wave.
For the time machine, we will interpret the interference pattern as bit 1 and no interference pattern as bit 0.

Now the DCQE:
This device lets us choose if we know the path of the particle or if we want wo erase this knowledge. And we can make this decision after the particle hit the screen (that is the "delayed" part), with the help of quantum entanglement.
How does it work?
Each particle send out by the emitter will pass through a crystal which will split it into an entangled pair of particles. This pair shares the same quantum state in space and time. If we know the path of one of the particle "halves", we also know the path of the other one. Remember the knowledge about the path determines if we will see the interference pattern. Now one of the particle "halves" goes directly into the screen by a short path. The other one takes a longer path.
The longer path has a switch that we can operate (this is the "choice" part). The switch changes the path that the particle takes so that it either goes through a detector or it doesn't, determining if it will contribute to the intererence pattern on the screen or not. And this choice will be done for the short path particle-half because their are entangeld.
The path of the first half particle is short, so it will hit the screen earlier.
After that happened, we still have time to make the choice for the second half, since its path is longer. But making the choice also affects the first half, which has already hit the screen. So we can retroactively change what we will see (or have seen) on the screen.
Remember this has already been tested and verified. It works.

The time machine:
We need enough photons to distinguish the patterns on the screen for one single bit of information.
And the insanely difficult part is to make the path for the second half long enough to have something practical.
Also, those photons need to stay coherent during their journey on that path and are not allowed to interact with each other.
We could use two mirrors, to let the photons bounce between them to extend the path (or the travel duration), but those need to be insanely pricise for reasonable amounts of time.
Just as an example, for 1 second of time travel, we would need a path length about the distance of the moon to the earth. And 1 second isn't very practical. To win the lottery we would need at least many hours.
Also, we would need to build the whole thing multiple times, one for each bit of information.

How to operate the time machine:
Turn on the particle emitter and look at the screen. If you see an interference pattern, write down a 1, otherwise a 0.
This is the information that your future you has sent you.
Repeat this process with the other time machines for more bits of information.
Then wait the time which corresponds to the path length (maybe send in your lottery numbers) and then (this part is very important) make sure to flip the switch corresponsing to the bit that you wrote down, so that your past you receives that info in the past.

I hope that helps :)

A software developer's experience life cycle:

0 - 5 years: attempt to replicate what your current senior is preaching, assuming that's the right way. Reading "Clean code" and preach it as gospel, even though you don't practice any of it.

6-12 years: gained the belief that you are better off coming up with solutions yourself, usually "sophisticated" and "elegant" which to everyone else (and also yourself a few years later) is an over-complicated inheritance ridden shit show. You have realised the "Clean code" movement is actually a cult but still believe code reuse is the holy grail.

13+ years: finally realized that simplicity and pragmatism is the most sensible way for most software development. Code is now readable, maintainable and functional. You took the few good bits from "Clean code" and ignored the extremism. These are the golden years.

The problem is most developers jump ship and stop developing before reaching the golden years, thus resulting in most software projects looking like shit.

Unpopular opinion, but it doesn't make it untrue.

People always complain about node_modules, at least it works most of the time, unlike god-damn python AI packages. I can't install https://github.com/shashikg/... because for some inane reason it requires a LLVM wrapper that in turn wants to compile LLVM manually. Not just manually YOU, the filthy dev that wants to use this holy library, must go and compile it yourself. And don't you forget to apply the holy patches.

God the code in AI projects always sucks - last time I looked at voice gen it required me to figure out what the fuck it was doing and go hunt down the libraries manually and still have to vendor in half the library because they used a library that had a minor migration but was 2 major versions behind or similar issue.

Why do science bros write such shit code that always wants to reach into things completely unnecessarily. Just write your shit properly.

Is this why they always have docker image...

This $1 mofo created more engineers than Harvard or Stanford.