Cool to see that EurIPS is now official as “NeurIPS Europe” but EurIPS is such a cool name

Reminds me of the YOLOv3 paper, which cites itself (not that it matters much, since it now has almost 40k external citations)

3/3 papers accepted to #ICLR workshops and 1/1 to the conference track. See you in Rio!

Arinbjörn Kolbeinsson Machine learning scientist at UVA and Imperial College London. Research in healthcare AI, foundation models and legal technology.

Annual tradition: my personal site runs virtual fireworks on Dec 31 / Jan arinbjorn.is

Glad you like it!

Routing data from signal.eu.org, station data and basemaps from @openstreetmap.bsky.social

Trainmap | Track Your Train Journeys Create beautiful maps of your train journeys across Europe. Track, visualise, and share your rail adventures.

Made a thing: an interactive train route mapper for Europe 🚃

Pick your stations, see the actual railway paths, share your routes.

No signup, no accounts, no friction.

trainmap.app

By train to Copenhagen | SBB Take a city break and travel by train to Copenhagen. Buy train tickets online.

Cool! EurIPS was just a few months too early for those of us based in Switzerland to use the new direct night train to Copenhagen (starts April 15): www.sbb.ch/en/leisure-h...

Back from #EurIPS Copenhagen. The European AI "style" came up a lot, which is less about where research happens, more about how: rigour, efficiency, interpretability over pure scale.

Hope it returns in 2026 (offset by a week so we can do NeurIPS Sydney too).

Phrase of the conference: Pareto-front

Uncertainty in the LLM era - Science, more than scale Today's AI narrative is anchored in scale, and models are far from textbook statistical modeling. And yet, the plague that are hallucination shows us th…

Uncertainty in the LLM era - Science, more than scale
speakerdeck.com/gaelvaroquau...

My #EurIPS keynote, on scale, uncertainty, and narratives in AI

Great to see the lossless version! Although I really like the uniqueness of the physical one

It’s starting to feel a lot like (N)EurIPS

@maksym-andr.bsky.social

Looking forward to an exciting week at the inaugural @euripsconf.bsky.social in Copenhagen

Within two years, the largest data centers may use more power than major cities.

We think Microsoft's Fairwater Wisconsin campus will have a peak electrical load of 3.3 GW when its fourth building comes online in late 2027.

(Los Angeles used 2.4 GW on average in 2023)

Contributing to the #aurora posts from Reykjavík

E.g.
Case where it helps
A) My parents, Ayn Rand and God (Ambiguous)
B) My parents, Ayn Rand, and God

Case where it hurts
A) The leader, JFK and Stalin (Not ambiguous)
B) The leader, JFK, and Stalin (Ambiguous – is the leader and JFK the same person or separate?)

The show example is a mix of both.

I'm not sure that's what happening here.

The Oxford comma clarifies whether the final two items are separate entries or appositives of the third-to-last item. However, it cannot disambiguate when the second-to-last can be an appositive of the third-to-last item.

I am still not seeing how people think adding a comma will reduce ambiguity here

That would make the sentence more ambiguous

It’s always a nice morning when you wake up to a paper acceptance notification

Mosquitoes arrive in Iceland - RÚV.is Three mosquitoes were found in Kjós in recent days. This has been confirmed by the Icelandic Institute of Natural History. It is the first time that mosquitoes have been discovered on Icelandic soil.

Iceland's status as utopia has come to an end.

No more mosquito-free haven.

www.ruv.is/english/2025...

Great conversations at Launch @ethz.ch today on what it takes to scale infrastructure plays in Europe.

The speakers highlighted critical new challenges that startups face, especially around timing, market entry and what differentiates category-defining companies from point solutions.

Sora 2 can solve questions from LLM benchmarks, despite being a video model.

We tested Sora 2 on a small subset of GPQA questions, and it scored 55%, compared to GPT-5’s score of 72%.

Very interesting work. I wonder if it could be a video-language fusion model, partly trained on high quality text

I’ve already used all of mine unfortunately, will let you know if they refill!

I was at a dinner party last month where this topic came up. Would recommend.

My journey into the artificial world of Sora 2 - Poynter OpenAI’s new social media platform is filled with AI slop and people scamming for likes

Good stuff from @alexmahadevan.com

Thank you!

Did you get any additional ones? I’m also hoping to find one