Quite interesting that I still haven't been able to find a person in Denmark who was positive about their working experience in Novo Nordisk *I usually hang out around AI/Robotics folks

God I hate LinkedIn

Are there some people from the Hugging face team who can help with the rate limit for new user ? We have a team member in Danish Foundation models joining the Dynaword project (open source datasets for Scandinavian languages) we would appreciate your help 🙏

Summer in Copenhagen:

You actually don't need a car to go to the beach

For some reason FSDP sounds like far right German party

How I imagine people working on LLM-assisted math proofs :

Oh well, so much for all of the "digital sovereignty" talks www.ft.com/content/681c...

I don't want AC, I just want a world without climate change

Almost all the stories about humanoid robots that I hear start with "that's really cool" and end with "we're waiting for details from China because the wrist/leg/(insert body part) broke"

Happy father's day guys !

I think it is called test-time compute because it is testing my patience

Being in the office in Copenhagen where 60% of the team are from Sweden is pretty funny experience

Sweet summer child.... Now I have started training VLAs and this list got way bigger

Hot take : The average school teacher is far more harmful to a curious and ambitious child than ChatGPT

We need C-drama about Chinese AI CEOs and we should call it "Pursuit of AGI"

Looking back on it, how Danish universities handled Mette's attack on international students last autumn was pretty disgraceful Literally same level of cowardice as with US universities and Trump

Unfortunately, it seems that I prefer doing things that bring 0 shareholder value

En se rendant à Erevan le spécialiste de géométrie complexe Mikhail Verbitsky a été arrêté par les autorités arméniennes le 12 juin sur demande de la Russie. Il risque l'extradition. La SMF appelle les autorités arméniennes à le libérer et appelle à signer la pétition www.change.org/p/free-misha...

Also do you remember there were these guys from "AI application layer" startups that advocated for building on top of frontier models APIs ? I am slightly worried about that business model rn 🙃

It is almost like we shouldn't rely too much on the American LLMs in Europe

Apparently now we can say the same for Anthropic vs US government lol

Political parties that advocate for "mobility freedom" (i.e "coexistence of car, bicycle, public transport and pedestrian streets") - are usually just one of the most car-brained populists imaginable *Inspired by reading the FDP programme in Berlin

So, the plan is this: We take all the research ideas, run them through Fable 5 - and only work on those that it rejected because of the safeguard

I have always thought that people didn't hate Anthropic enough This whole thing about "being the only lab who is responsible enough to frontier AI models" is really pretentious and seems to be evolving into "we are the only lab who is responsible enough to use frontier AI models"

Anthropic employees reading social media after the release of Fable 5 :

That is the stupidest thing that I have read during this week

I mean they had really good blue trams though

The problem with VC funding is that you will eventually run out of letters in English alphabet to name a new funding series

During the witch hunt against international students that the Danish Government initiated last autumn I actually googled what kind of university level education Christina Egelund (Minister of Education and Research) obtained And guess what....

You either die as a frontier AI lab, or live long enough to become a neocloud

Freedom to speed = freedom to kill It is not whose inconvenience counts, it is "do you value human lives or not". If you do - there is no question why car traffic should be heavily regulated in densely populated areas and why we should aim for VisionZero

We've come a long way, I can play chess with NeurIPS paper online maiachess.com - human conditioned engine, pretty instructive to see how model thinks people with different rating would play in certain positions

I am starting to think that empathy is disqualifying trait for AI labs

Social media be like: > Someone writes a post > People hallucinate context they don't know about the author/situation etc > React to the post according to their hallucinated context These same people then will complain about LLMs lol

I think there a joke in urbanistic community that Sweden was neutral during WW2, but Stockholm decided to bomb itself afterwards

My conspiracy theory is that Trump understands that they will lose midterm and next election, so he want to hurt country long term so the economical crisis will hit during Democrats government Or (more likely) He is just dumb

But in all seriousness: Another dump policy by Trump government that EU, no doubt, will completely fail to capitalize on (same as with research funds cuts) One more reminder that we need United European government with real power so we can quickly act in situations like that

Oh so that is why Anthropic opened office in Milan

What was the record for the government formation I Netherlands, like 300 days for some Rutte cabinet if I remember correctly So still a long way to go

0 days without car/airline industry bailout

Agreed, I also think work and research on data attribution it really important for the future of AI in science (and in general our lives)

This OpenAI proof kind of makes me a little bit sad that I haven't continued with Mathematics as my Masters after Bachelor in Lund Although now I am objectively much happier and education seems to suit me better - I just think it is super exciting time to be a mathematician

Just realised that European Union is literally a MoE model

Urbanistic hot take : Jaywalking should not be a crime and people should be able to SAFELY cross any road in the city at any given time at any place Because cities are for the people and not for the cars

Well that was fun

For me bitter lesson always was somewhat confusing, clearly he speaks to ML researchers not using learning based methods instead of hand crafted heuristics But the thing is : Learning based methods are actually arguably even more dependent on human knowledge (because of the training data)

Max Verstappen is another evidence that Sim2Real is working

My top 3

I think it is a good initiative, but I would like to see more transparency on how it will be implemented and whether it will be possible to have some appeal system ->