The goal of the hackathon is to introduce people people to the Dynaword and me and @kennethenevoldsen.eurosky.social will be there offline and online to help with your first contributions and answer your questions about the project

We are organizing Dynaword mini-hackathon

Danish Dynaword is the largest continuously developed collection of openly licenced datasets for Danish language. We are currently expanding it to Swedish, Norwegian and Icelandic and looking for new contributors !

1/2

Anthropic got compute problem and that's why they didn't release Mythos

Really makes them feel more human and relatable (I am also GPU poor)

Great stuff

This is the company behind the software, used quite a lot of different sensors

www.adastec.com

Belgrade cats

Also it seems working on data as a community is the only way we can meaningfully provide alternative and perhaps catch up performance-wise to closed source models

* I am biased since I work on Dynaword with @kennethenevoldsen.eurosky.social

It seems that we are slowly moving away from traditional static datasets and start treating data as additional axis of model development and open source ecosystem

Yeah maybe we are quite early and making many mistakes, but I think it is a direction worth pursuing

I am so freaking tired of so called "urban experiments"

When new road or even highway is being built do we get a trial ?

> No

When new car parking is being built do we get a trial ?

> No

So maybe when we finally get human-friendly space we can introduce it without a trial PERMANENTLY ????

LoMa-R is our newest addition to the LoMa family. In our IMW paper at #CVPR26, we investigate rotation invariance in the sparse matching pipeline. The resulting model is robust to rotations, even matching star constellations, and achieves strong upright performance.

github.com/davnords/loma

Sorry if is sounded a bit toxic

I just think that Europe is a major innovation hub, we should think like a major innovation hub

I totally disagree with the whole "we can't compete" crowd

I find it extremely funny :

> Swedish "AI application bros" criticizing Swedish government investment in the foundation models/data stating that "dOn'T coMpeTE wiTh tHe US oN foUnDatiOn MoDelS, bEcOMe tHe bEsT At aPplyIng TheM"

> proceeds to get replicated in the next Anthropic release

Yeah exactly, like you can input past lecture notes for the context and the ask it to generate some questions and It will give you GUI with 4 choices where you can click

I usually prefer also to guide it what exact topics I want to practice and maybe range the difficulty

And this creates some kind of stale mate situation, when I am trying to choose whether offload some part to accelerate research or dive deeper, much slower but getting more out of it more my brain

Gemini interface to multiple choice is quite cool though, but I think every major LLM has something like this know

I very much agree about cognitive offloading, but I also even more worried sometimes that I am inefficient and gonna be left behind because I under-utilize the LLMs ->

I also like to "derive" something with chat GPT, like I pick the topic and trying to go through the problem explaining solution to chat GPT, if I am stuck I can ask for hint or direction I should think

One can argue that this is bad considering it removes friction from learning, but imho it is net-good because it will allow more people to understand certain subjects like mathematics, where classical literature is sometimes very poorly written

I think it is more about the fact that learning became more personalized, like before chat GPT I needed to go through 10 analysis books to get to the explanation I need, but know I can ask ChatGPT to express it in the way and using the analogies I understand

This is an important win, but

1) last time Orban sat in the opposition for 8 years and came back even stronger

2) Magyar is very conservative right and seems to share Tramp-like rhetoric (www.oxfordstudent.com/2026/02/20/p...)

But I think now we can celebrate a bit and worry about that later 🥂

Hopefully next time will visit Budapest with Orban gone for good !

Btw on a second photo you can see "Children railway", narrow gauge line run by children

gyermekvasut.hu/en/home/

That being said if I understand correctly Orban have cut funding for the public transport for example and in general tried to make life harder

I mean Budapest itself doing pretty good I think, they have quite likable mayor from the opposition and even received some EU funds bypassing Orban government

Yeah, very stylish metro line and park also

Spending this weekend in Hungary

Budapest has some of the most diverse and beautiful public transport fleet I have ever seen

From modern Caf urbos to art nouveau era trams to Cogwheel Railway

It'll be funny that in the end Anthropic's competitive advantage over OpenAI will come down to the fact that their CEO knows what gradient descent is

Ahahaha, yeah I am afraid exhaustive version will lead to recidiv of my depression

ML Researcher's Dilemma:

The model performs poorly because

a) it's crap
b) the benchmark is crap
c) the data is crap

But for some reason, you always end up concluding that the crap one is you

Yeah, but I was for some reason thinking that is maybe some slight disadvantage to show it, since competitors might speed up/move more resources to catch up

To be honest this was pretty obvious to predict, but I am quite surprised they have showed the benchmarks for it

Maybe to improve hype around the company, given that they consider IPO ?

Lund, Aarhus, Odense, Graz, Utrecht, Bergen

I am a sucker for small-medium sized walkable "university" towns in Europe that have trams