How Europeans view average American city :

Also realistically Chinese labs most likely aren't directly training on Claude's outputs (because not 150k, not even 13 million is enough for something like a frontier-level performance lol) My prediction - most likely they're using Opus outputs as seeds for synthetic data

Ok now I want to distill Opus just out of spite

Nah they should stop pretending this is some "national security" issue They have cut access to xAI and OpenAI for their coding tools also. It is pretty obvious that Anthropic is afraid of the competition at this point techcrunch.com/2025/08/02/a...

I must admit, that at this point I would strongly benefit from the university course that focuses just on model evaluation (both CV and LLMs) Like I am taking many courses on how to train stuff, but imho it is as important to be able to evaluate what you have trained

Sweden announced new "AI strategy" I think it is quite cool that more and more Nordic countries are interested in training open Language Models news.cision.com/knut-och-ali...

I wonder what if doing speculative decoding this way : Take smaller/distilled/heavily quantized model and run it locally, then generate "proposal" tokens for larger target model that runs on the cloud

When you have stopped being GPU poor and realized there is whole another PyTorch to learn

It would be nice to see "cost" axis, I think this might be actual moat of open weights/open source models

Life update: In March, I’ll be starting as a Student Developer on the Danish Foundation Models project at Aarhus University I am excited to work with amazing people and contribute to Danish open source 🇩🇰

I totally support this message : lichess.org is great project, with a lot of cool features + fast and smooth interface + you can actually play for free without watching adds on chess website (crazy I know)

I think it is very good analogy also because lets say F1 drivers actually learn the track prior to competing (during simulator runs and practice), so they have very good mental model before start pushing to the limits

Next news : Gemini Pro teams sued Gemini Flash team over intellectual property theft

"The company considers distillation to be intellectual property theft, it said" AHAHAHAHHAHAHAHAHAHAG,LMAO

I love Norwegian approach to open source LLMs Just make AI-reseach team IN THE NATIONAL LIBRARY huggingface.co/collections/...

"$2 million in licensing costs annually supporting 100,000 users." Imho spending $2 million (ANNUALLY) on ChatGPT during literally worst cuts in American science is borderline criminal

I might be wrong, but from my very limited experience I feel like AI researcher day is ultimately consists of waiting for GPUs to become available and then waiting for training to finish

Reminds me when putin called ex-YANDEX CEO "talented man" and next day guy hired bunch of security

Arguably training network doing people recognition might be even easier then doing dog recognition for them

Actually I might do the list of where to eat/what to do in Malmö/Lund during ECCV I think Southern Sweden quite underrated and often overlooked due to Copenhagen proximity

If I am learning but my network is not, can it still be count as a success ?

Random realization at 4 am : It seems that during my bachelor in mathematics I have managed to train my brain to compile latex in my head

"Norway is already falling behind" is crazy guys. I understend need to hype up the investment into competition, but afaik Norway is doing way better then Sweden, Denmark and Finland combined (at least from Foundational model stand point)

The knight of the seven runtime errors

Executives trying to figure out how to replace software Engineers :

I think we should invent new swear words The old ones don't cut it to describe current events accurately

It is a shame that wandb repo is not released together with training code I think would be nice for transparency + reproducibility

I am sure all dictators/authoritarian leaders in the world are very happy

3D Computer vision people

Now that is a proper winter

Unironicaly there is a book for kids called "Warranty Men" from Eduard Uspensky - about little people who live inside devices and mechanisms and repair them during the warranty period. And by this book cartoon was filmed

Also good thing to keep in mind, that afaik the current "subscription token economy" is heavily subsidized and I am not sure Anthropic/OpenAI are charging necessarily full amount (i.e the same price that would charge for API cost)

I have seen such case also : >Dataset released in ICLR 2025 > In January 2026 you can't access dataset because it was deleted from Hugging face (likely due to licencing issues related to using stuff from Common crawl)

I still can't calibrate myself about Danish robotics ecosystem I have heard so many times that Odense is European capital of robotics, but I haven't heard about any startups from there

What if position encodings were designed for vision from scratch? We introduce PaPE—Parabolic Position Encoding. Outperforms RoPE on 7/8 datasets and extrapolates to higher resolutions without fine-tuning or position interpolation. Paper, code, and website in thread 🧵

It is so energy draining/frustrating + extremely anxious to be stuck in the limbo because you are waiting for a decision out of your control

Went to Vienna last weekend and the public transport was really incredible Might be one of the smoothest experiences I have ever had

Nothing motivates you to complete all objectives of your space mission faster then a Grok-based voice AI assistant on board

I don't want to defend Trump voters, but I don't quite like "sane nation" sentiment : In Britain Nigel Farage is popular again, AFD getting popular in Germany, Finland/Sweden "ellected" far right during last election (as part of government coalition, but still) and so on ->

I couldn't connect the dots for quite some time but finally figured what current US situation reminds me of (i.e likable mayor of major city vs corrupt president) This is actually almost like Gergely Karácsony (mayor of Budapest) vs Viktor Orbán (prime Minister of Hungary)

I don't want to be rude, but imho it is not "AI noticeably degraded programmers" it is more like "Programmers that used AI to substitute their thinking process degraded themselves"

Took a break from coding with one Claude to watch another at Albertina

I agree but also I think it is important to mention that there were a lof of AI in our phones/browsers long before even GPT-2 But it was framed as just "useful new feature" and not force fed undercooked product under shiny wrapper

Anxiety disorder 101 : You were anxious about something, forgot about it for a split second and now you are worrying about what was that worry you forgot And this is how you get anxiety squared

It is almost like cutting down/calming car trafic is more efficient solution then self-driving cars

"Slow horses" is a great series imho As someone who watched probably too much of "2000-2010 superhuman genius FBI detective" type of stuff - this one was nice refreshment and I kind of like British humor

Optimized sleep loss function too hard and now I have to somehow survive until 2 pm

My mother was born in Ukrainian SSR, my father was born in Kazakh SSR So my first complement I have heard from GPT-3

I hate point clouds

This country plans to host World Cup btw