Sign in

Vésteinn Snæbjarnarson

@vesteinns.bsky.social
943 followers 859 following 1 posts

PhD student at the University of Copenhagen Visiting researcher at ETH Zurich '24-'25 ELLIS student doing NLP, CV, ML vesteinn.is

PostsRepliesMedia
Reposted by Vésteinn Snæbjarnarson
the babahuck @tylerhuckabee.bsky.social · 17/09/2026
god this is good
flight of the concords scene with murray, brit and jemaine. 

"How's this, then? Ed Sheeran, beloved by all, left without an opening band at present. Enter you two."

"What if we want a free Palestine, Murray?"

"I reckon you'll get lots of things for free when you're proper rockstars. This might be your ticket."
375082956
Reposted by Vésteinn Snæbjarnarson
Ásgeir Berg @asgeirberg.bsky.social · 14/09/2026
I thought this was a good time to mention (again) that @fennert.bsky.social and I wrote a paper arguing that using LLMs to generate and verify proofs does not necessarily give us mathematical knowledge as that notion has traditionally been understood: www.mxphi.com/all-issues/v...
mxphi.com
The Philosophical Prospects of Large Language Models in the Future of Mathematics
The Philosophical Prospects of Large Language Models in the Future of Mathematics — Fenner Stanley Tanswell, Ásgeir Berg
133
Reposted by Vésteinn Snæbjarnarson
Ásgeir Berg @asgeirberg.bsky.social · 02/07/2026
Two things that foreigners don't understand about Icelandic whaling are (a) it is a way for Icelanders to perform their patriotism/nationalism. Without foreign activists trying to stop it, it would have stopped a long time ago, and (b) it is a way for Icelandic conservatives to own the libs.
152
Reposted by Vésteinn Snæbjarnarson
Sebastian Loeschcke @sloeschcke.bsky.social · 03/06/2025
Check out our new preprint 𝐓𝐞𝐧𝐬𝐨𝐫𝐆𝐑𝐚𝐃. We use a robust decomposition of the gradient tensors into low-rank + sparse parts to reduce optimizer memory for Neural Operators by up to 𝟕𝟓%, while matching the performance of Adam, even on turbulent Navier–Stokes (Re 10e5).
2317
Reposted by Vésteinn Snæbjarnarson
Srishti @srishtiy.bsky.social · 02/06/2025
I am excited to announce our latest work 🎉 "Cultural Evaluations of Vision-Language Models Have a Lot to Learn from Cultural Theory". We review recent works on culture in VLMs and argue for deeper grounding in cultural theory to enable more inclusive evaluations. Paper 🔗: arxiv.org/pdf/2505.22793
Paper title "Cultural Evaluations of Vision-Language Models
Have a Lot to Learn from Cultural Theory"
35818
Vésteinn Snæbjarnarson @vesteinns.bsky.social · 25/01/2025
When I grew up in Sweden in the 1990s the government distributed the book “Tell Ye Your Children”. Its purpose is to remind people how fascism, and seeing oneself as better than others can lead to inhuman evil. I think about it often and recommend you read it too.
levandehistoria.se
0124
Reposted by Vésteinn Snæbjarnarson
Belongie Lab @belongielab.org · 24/01/2025
SCIA 2025 submission deadline extended to Feb. 17
scia2025.org
SCIA 2025
Scandinavian Conference on Image Analysis 2025
2215
Reposted by Vésteinn Snæbjarnarson
Belongie Lab @belongielab.org · 03/12/2024
LoQT: Low-Rank Adapters for Quantized Pretraining: neurips.cc/virtual/2024... Authors: @sloeschcke.bsky.social, @madstoftrup.bsky.social, @mkastoryano.bsky.social, @serge.belongie.com, @vesteinns.bsky.social 🗓️ West Ballroom A-D #6104 - Thu 12 Dec 4:30 p.m. PST — 7:30 p.m. PST (2/5)
neurips.cc
NeurIPS Poster LoQT: Low-Rank Adapters for Quantized PretrainingNeurIPS 2024
141
Reposted by Vésteinn Snæbjarnarson
Shauli Ravfogel @shauli.bsky.social · 12/11/2024
Happy to share our work "Counterfactual Generation from Language Models" with @AnejSvete, @vesteinns, and Ryan Cotterell! We tackle generating true counterfactual strings from LMs after interventions and introduce a simple algorithm for it. (1/7) arxiv.org/pdf/2411.07180
2143
Reposted by Vésteinn Snæbjarnarson
antonio vergari ⚔️ short-circuiting @nolovedeeplearning.bsky.social · 20/11/2024
welcome @ellis.eu! could you please extend the deadline to review #PhD applicants? 🙏 as things are now, you are giving (postdocs, senior PhD students) reviewers 5 working days to evaluate 16 (and more?) applications... otherwise reviewers will do cheap evaluations looking at the usual "metrics"
0144
Reposted by Vésteinn Snæbjarnarson
Marvin Schmitt @marvin-schmitt.com · 19/11/2024
Let’s welcome @ellis.eu to Bluesky and give them a follow! 🦋
0367
Reposted by Vésteinn Snæbjarnarson
Sebastian Loeschcke @sloeschcke.bsky.social · 18/11/2024
Excited to see LoQT adopted in memory-constrained settings, lowering the barrier to training models such as LLMs! Great collaboration with @mabeto5p, @mjkastoryano, @sergebelongie.bsky.social , @vesteinns.bsky.social Code: github.com/sebulo/LoQT 💻 Paper: arxiv.org/abs/2405.16528 📄
github.com
GitHub - sebulo/LoQT
Contribute to sebulo/LoQT development by creating an account on GitHub.
131