It is actually incredible that most likely USA will achieve AGI before actually paying their service industry workers a livable wage
Chris Paxton@cpaxton.bsky.social · 2d agoYikes
Vladimir Salnikov
@v4ldelund.bsky.social
Currently studying Masters in Human centered AI in Technical university of Denmark, interested in 3D vision and Vision Language Models Previously bachelor student in Lund university
Chris Paxton@cpaxton.bsky.social · 2d agoYes actually that is very good point, compare recent spam attack vs how they published Unit distance proof - nice popular science blog, video, paper and accompanying papers (much better written with clear references and acknowledgements to human ideas and part work that lead to the breakthrough)
Ethan Mollick@emollick.bsky.social · 2d agoIts reasonable for mathematicians to point out that more proofs are not always the same thing as advancing mathematics. But it suggests a need for new goals for what math is trying to do. I think the same thing will happen everywhere. 100x more PowerPoint or code is not always progress - what is?
How I imagine Linear Algebra professors right now telling students that matrix multiplication is about to solve mathematics :
Robert Downey Jr. as Tony Stark in desert scene
ALT: Robert Downey Jr. as Tony Stark in desert scene
static.klipy.com
Meanwhile in AI conference in Denmark (It will either save my mental health or will completely destroy it)
Eugene Vinitsky 🍒@eugenevinitsky.bsky.social · 4d agoOnce again, if you are a senior researcher spreading doom to junior researchers, shut up shut up shut up. It’s your job to figure out a path forward, even in high uncertainty times. Come back when you have something constructive.
Math Faculty expenses before 2026: Chalk, blackboards Math Faculty expenses after : Chalk, blackboards, a ChatGPT Pro subscription for the entire faculty
On one of my uni courses, they give extra points toward the grade for completing mini-projects, but you have to post them in the Microsoft Teams chat I decided that I don't need those extra points so badly to willingly use Microsoft Teams......
While OpenAI release is quite impressive... There gotta be a better way to share math research the dumping it like a slop on GitHub, like what are we doing seriously
Though it is not really I agree that one can use open weights model like Qwen/GLM/Kimi/Deepseek that captures most of the pre-training data sufficient for above average local intelligence But the problem is that the data itself is not accessible - one can not *legally* train really good LLM
Ted Underwood@tedunderwood.com · 6d agoSomething not sufficiently registered in the convo around AI/IP: the value that was gained from pretraining (on our collective legacy) is now common property. People do pay more for frontier models than open ones — but the differences they're paying for, at this point, come from post-training.
Well www.nyidanmark.dk/en-GB/News-F... It was kind of expected considering pre-elections talks but still unbelievably idiotic decision
SIRI: New rules for third-country students in state-approved higher educational programmes
On 1 October 2026, new rules will come into effect for certain third-country students in state-approved higher educational programmes in Denmark. The rules will also affect students’ ability to bring ...
nyidanmark.dk
Vladimir Salnikov@v4ldelund.bsky.social · 2mo agoI understand why he says that, but objectively every action of Danish government points that it wants 0 international students in Denmark Perhaps long term return on having international talent is much harder to measure then short term gain from swinging few far right voters
When you look at the professors website trying to understand cold-emailing instructions :
My radical suggestion - if data for the model is not open source we should assume by default that benchmarks are contaminated and model performs worse then numbers suggest
The thing that drives me crazy - there are probably >1B fairly open tokens for Faroese out there (based on my scraping of the government/municipality websites + data from scientific journals etc) But we just can't release it, because majority of it is not covered by copyright exception law
Good thing that I was never smart to begin with
Vasilii Vadimov@vvadimov.bsky.social · 3w agoI physically feel how vibe coding makes me dumber
Cyclists in Copenhagen the moment you need to cross a bike path :
Max Verstappen: Victory Moments
ALT: Max Verstappen: Victory Moments
static.klipy.com
Sorry guys, but "System 1 Decision model" is something on claudish Can we just call it classifier ?
I think conferences should introduce something like Golden Raspberry Award but for worst AI-slop paper
📜CALL FOR PAPERS📜 We are looking forward to your submissions to NoDaLiDa 2027, to be held in Copenhagen on May 25–28. Submission deadline: January 18, 2027 (direct submission), February 22, 2027 (ARR-reviewed papers). eventsignup.ku.dk/nodalida-27/...
The 26th Nordic Conference on Computational Linguistics (NoDaLiDa)
Five-day conference on computational linguistics, natural language processing, speech and language technologies hosted by the University of Copenhagen (UCPH).
eventsignup.ku.dk
Unique opportunity to advertise huggingface.co/danish-found..., pre-trained from scratch on quite interesting architecture and punches above its weight class
danish-foundation-models/DFM-Mimir · Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
huggingface.co
Knut Jägersberg@knut-jagersberg.wsocial.eu · 3w agoit is worrisome russia can put out an increasing number of their own pretrained foundation models, whilst european output shifts towards fine tuning chinese models. huggingface.co/yandex/Alice...
Judging by internet discourse in the future we will end up with 2 programming languages: 1) Rust 2) Lean
ai.escience.sdu.dk We will generate enough synthetic data to make Scandinavian data great again
UCloud AI: Sovereign AI inference for Danish research
Run large language models on your own data, inside your UCloud project. Built by the universities, running on Bitten. Every token computed in Denmark, with energy-efficient computing and heat recovery...
ai.escience.sdu.dk
Afaik at this point EU had two solutions : 1. Treat this situation like WW2^covid, mobilize resources and try to catch up at least with China on AI & robotics 2. Do nothing And this is essentially the second option on that screenshot
Alexander Doria@dorialexander.bsky.social · 4w agoyeah we’re cooked (von der leyen state of union)
Literally all AI companies public comms can be described with this meme :
David Picard@davidpicard.eurosky.social · 4w agoSo now it's "AI may kill us all, we have to stop" Next month will be "Our latest model is so powerful that we have to restrict its access, sorry guys!" In three months: "You can now access that super powerful model! Try it, it's cheap!" In 5 months: "big model token cost go brrr" Rince, repeat
I think I have reached the final boss of ambient study music on YouTube
Actually researchers at Faroe Islands done great job at collecting data over the last few years and we will likely have more then enough tokens to generate synthetic instruction data and add that post training mixes in DFM projects (not that SF tech bros care for low resource languages)
Huh, apparently Talgo carriages began to appear on domestic IC routes in Denmark (I hope I can get to uni before this thing breaks)
