Trying to filter out multilingual instruction tuning datasets and this meme is the only way I can describe the situation

Diving into LLM (post) training in (almost) 2026 is a fascinating pursuit Just this kind of vibe of clueless kitty, when you're trying to figure out what the hell happened over the last two years

Also I have created a collection with image-text datasets for Danish on Hugging Face Hopefully next year will expand it for other Scandinavian languages huggingface.co/collections/...

As a small Christmas present I have created the Image-text dataset using digitalized artworks from National Gallery of Denmark and their text descriptions in Danish/English (when available) The dataset was collected using SMK API (www.smk.dk/article/smk-...) huggingface.co/datasets/V4l...

Enjoying the perks of finally having something to show from my studies beyond proofs in Latex This semester during the deep learning course we have made a Few shot Anomaly detection model ->

Last week I spent 4 days volunteering at EurIPS It was an amazing experience overall and it was great to attend my first "AI conference" here in Copenhagen

Image-caption dataset, but instead of describing the content of the image people will describe what they feel looking at the image

It is interesting that synthetic data is popular within Optical flow research for quite some time already (at least my impression), but only recently started to propagate to other vision tasks (like monocular depth estimation)

Learning about synthetic data generation for computer vision and encountered Flying chairs dataset I have so many questions, I don't even know where to start

They really did put Open in the OpenReview

I am strongly considering doing a PhD after my masters, because I feel it is something that might suit me (more ownership/personal freedom/ independence and I feel I like discussing and doing research in general) But I am so afraid I am not good enough for it ->

Denmark in November - outside looks like grayscale Depth map

My GPU :

The fact that Denmark still doesn't have electrified railway till the second largest city - is kind of shocking Hopefully it will be fixed soon, and train rides will become more enjoyable and sustainable

I am 99% sure that physically based rendering was created so that physicists could feel on their own skin the same pain that mathematicians feel looking at physicists doing math

What is the difference between "Inverse Rendering" and "Computer Vision" ? Or is it just "Inverse Rendering" is what graphics people call computer vision, like how some people in the vision call graphics image synthesis ?

Realistically predicting the AGI timeline is like waiting for the Deutsche Bahn train : Every time the expected departure time just keeps moving further until it is either cancelled, or just wait long enough and pick a next train

In line with recent trends I will create my own AGI timeline and call it : AI n+1

reviewer 2 - lawful good version

Why can't it be like this for the whole winter 😭😭 I hate global warming

Great thing about my education: I can take up to 30 ECTS in so called "special courses", which are just research projects under PhD student supervision Which is incredibly fun and way more exciting then regular courses (well at least for me, I am not sure about PhD students who supervise me 😁😁)

Before the new Olmo 3 release I thought I was not colorblind Now I am questioning that

Only SAM I care about :

Just realized that right now on my PC model called "Baguettotron" runs on a benchmark called "Common sense reasoning" in Croatian

Copenhagen municipal elections in a nutshell:

I am wondering if it is possible to use synthetic data for VLM training (maybe as a supplemental part to real world image-text pairs) Haven't seen any releases on this topic

Got an email that I was accepted as a volunteer at EurIPS 🎉🥳 Explained to my girlfriend that EuroIPS is like a Comic con, but for those who know what gradient descent is

Hear me out : University course in AI where students will write tech blogs and post them on their web page, instead of regular notebook-style assignments

My first bike in Denmark got stolen 🎉🎉 I mean partially it was my fault (I left it near the S-tog station for a couple of days) But I honestly thought that my shitty 100€ bike won't be worth the trouble (clearly I was mistaken)

There is a new night train between Basel and Malmö 🎉 Pros : 1. New night train in Europe 2. Quite affordable at the time of ticket release (seated around 40€, bed in shared compartment starts from 150€) Cons : 1. The train goes through Germany (meaning no way it will arrive on time)

I am not so sure about the construction cost, considering how damaging the bus of this size to the asphalt below it When I was living in Lund I saw that almost every year the municipality was changing asphalt near bust stops because of the damage that even regular Man Lions City were doing

Do I understand correctly the trend that almost every new model in Computer vision is at least "Foundation model" And if it is generative, then it is "World model" ?

Overall I feel like I am a hobbit that left the Shire (Denmark) and went to Misty Mountains (Norwegian fjords)

Travelled beyond the arctic circle (to Tromsø), saw Aurora two days in a row already 🥳

I always knew that politicians are not required to have a relevant education to work in some specialized Ministries (like Energy/Agriculture etc) But being a Minister of Higher Education and Science without having a higher education is borderline comical 🙃🙃

Today I took a photo with a newspaper TO PROVE TO SWEDISH MIGRATION AGENCY THAT I HAVE LEFT SWEDEN SO THEY CAN CANCEL MY APPLICATION If I don't do that they will issue an expulsion decision (make it make sense please 🫠)

I actually got introduced to the concept of inputting "text" as "image" from this paper : arxiv.org/abs/2207.06991 and subsequent papers from Desmond Elliot who is working on Pixel language models in University of Copenhagen

AGI is when models would be able to transcribe this :

Turns out Fisher random is very cool ! You can get all sorts of interesting positions (and I don't need to learn openings at all)

Found computer vision paper that starts with : "In 1957, computer scientist Russell Kirsch and his colleagues at the U.S. National Bureau of Standards made history by scanning the first digital image. This pioneering work introduced the concept of the pixel...."

cphpost.dk/2025-10-01/b... Nah this is just pure racism at this point Disgraceful

I think one interesting problem I encounter being a Masters student during LLM-age is how to determine what knowledge requires deep-level understanding (and for example doing things from scratch) and what requires only a superficial understanding and can be automated using LLMs ->

I think we might have a way to convert all kids that like astronomy to computer vision

Sorry but this is such incredibly stupid and unfair decision cphpost.dk/2025-09-29/b...

Sunset and Copenhagen Metro

Expectation: AI companies will train new models to achieve human-like intelligence. Reality: AI companies will dumb people down through AI generated slop to the level of models. Well, overall, it's also a valid path to achieving AGI (and probably cheaper, too)

Decided to take courses in Rendering and Graphics at Uni to potentially use knowledge of CG in vision (synthetic data generation for 3D and other cool stuff) Currently trying to understand how to draw a triangle 🫣

To be honest this was an eye opening experience. I had some idea that curating datasets is hard, but I didn't know it was THAT HARD I probably made 100 mistakes along the way, but it was worth it The next goal is probably to create a VQA benchmark for EuroEval and QA datasets to VLM training

Hi everyone! We (Me and @kennethenevoldsen.bsky.social ) have created the first large image–text dataset for Danish that goes beyond captions by incorporating more context for each image (full-text Markdown and the nearest paragraph) ->

I don't think any EU official has done as much for the development of the European tech sector and science as Trump has over the past year lol