I don't know why, but for some reason, I really like seeing my CPU and GPU hard at work :)
PhD student at #INATECH @unifreiburg@bawü.social and #FraunhoferIPM, hobby graphic-designer, programmer, geocacher
♥️ Geodesy, track and field, vector art, music, travel
And yes, I created the profile picture myself 😍
I don't know why, but for some reason, I really like seeing my CPU and GPU hard at work :)
I am trying to use the new #DeepSeek #reasoning models for some programming help and within one day of using it, I managed to...
- leak system prompts through the thought process
- trigger indefinitely long thought processes (I had to kill the AI engine because it wouldn't stop thinking)
- Create a glitch prompt (similar to the SolidGoldMagikarp thing with #ChatGPT )
And I didn't even try to break it (I swear!). I really just wanted to use it for normal productive tasks!
Installing #Windows on a #VM be like:
1. Download a _fresh_ Windows 11 zip
2. "Installing"................. 99% (ah almost done)
3. "Preparing Windows"....(?)...... 86%
4. "We are preparing some things for you"... (no progress bar anymore !?)
5. "We're still working on it".... (I sure hope so!?)
6. "Installing updates" (remember: I just downloaded a fresh Windows ZIP 🙄)
Key changes: Fixed that flaw
Me: Noo, all you did was to make the code more ugly, but it still contains that very same critical flaw 😡
Sometimes with #AI it's really garbage in --> garbage out
Classic #ChatGPT :
Me: Please generate some #Kotlin code to fulfil this task.
ChatGPT: Sure, here's some Kotlin code that fulfils this task:
fun codeThatLooksLikeItFulfilsTheTaskButActuallyHasACriticalFlaw() {
}
Me: The code that you generated has this critical flaw. Please generate code that fixes that flaw.
ChatGPT: You're correct, my previously generated code had this flaw. Here's some code that fixes this flaw:
fun codeThatIsEvenMoreObfuscatedButStillHasThatSameCriticalFlaw() {
}
"Wir haben den Bahnhof Stuttgart mit 5 Minuten Verspätung verlassen. Grund dafür ist die längere Haltezeit am Bahnhof" 😂
An interesting case of #irony : Today, we visited the #mocomuseum in #Amsterdam . The #museum features some pieces of #art that criticize modern #consumerism and #capitalism and yet, at the end of the museum tour, we had to walk through a museum shop where we could buy those very same motives printed on t shirts 😂
For the first time, I can finally say "I've added AI to our product"
- Yes, it's actually useful
- No, I have not trained the model myself
- And yes, the model runs locally (I am not transferring 40+ TB of data to a cloud service 🙃 )
Das BSW hat noch andere Vertreter als die Wagenknecht?
Und ein Rambo Zambo im Konrad Adenauer Haus... Man muss den heutigen Tag doch mit Humor nehmen
Die AFD will also ein Verbrennerverbot durchsetzen. Ich wusste nicht, dass ich heute Abend doch noch eine Komödie sehen würde 🙃
@fleischie28 thanks 😉
It's always so much "fun" to debug a networking bug in a multi-threaded system with multiple pieces of software written in different programming languages all talking over the same bus....... 🤮
Wenn #IoT TCP auf Seriell-Konverter Motivationssprüche in siner GUI hat...
Today, I finally found a useful application of #AI: The Computer Vision Group at @unifreiburg created an AI that automatically scrapes the dishes served at surrounding #canteens and generates an image for each dish: https://lmb.informatik.uni-freiburg.de/essen/ I promise you, though: "Verschiedenes Vegetarisches" did not look remotely as nice as the AI predicted 🙃
#imagegeneration #generativeAI #food #university #computerVision
Luckily, all of this was only happening on a feature branch and I fixed the performance issue before it caused any real harm
TIL: By default, the H2 database vacuums the entire data base whenever a database connection is closed. In my case, I am using Kotlin Exposed to access the database which (by default) uses a new connection for every query. This meant that I was vacuuming the whole database after _every_ _single_ _query_ 🤦♂️
Using a connection pool instantly fixed this and reduced my test execution time from 60+ minutes to under 5 minutes 🙃