Excited to share our paper Representational Difference Explanations (RDX) was accepted to #NeurIPS2025! 🎉RDX is a new method for model diffing designed to isolate 🔍 representational differences. 1/7
1/8 🧵 GPT-5's storytelling problems reveal a deeper AI safety issue. I've been testing its creative writing capabilities, and the results are concerning - not just for literature, but for AI development more broadly. 🚨
My Lab at the University of Edinburgh🇬🇧 has funded PhD positions for this cycle!
We study the computational principles of how people learn, reason, and communicate.
It's a new lab, and you will be playing a big role in shaping its culture and foundations.
Spread the words!
🚀 DinoV3 just became the new go-to backbone for geoloc!
It outperforms CLIP-like models (SigLip2, finetuned StreetCLIP)… and that’s shocking 🤯
Why? CLIP models have an innate advantage — they literally learn place names + images. DinoV3 doesn’t.
I wrote a short rant about what irks me when people anthropomorphize LLMs:
addxorrol.blogspot.com/2025/07/a-no...
📢I am hiring a Postdoc to work on post-training methods for low-resource languages. Apply by August 15 employment.ku.dk/faculty/?sho....
Let's talk at #ACL2025NLP in Vienna if you want to know more about the position and life in Denmark.
New paper hot off the press www.nature.com/articles/s41...
We analysed over 40,000 computer vision papers from CVPR (the longest standing CV conf) & associated patents tracing pathways from research to application. We found that 90% of papers & 86% of downstream patents power surveillance
1/
"Researching and reflecting on the harms of AI is not itself harm reduction. It may even contribute to rationalizing, normalizing, and enabling harm. Critical reflection without appropriate action is thus quintessentially critical washing."
Fallacy of the Day:
Calling two different things by the same name doesn't make them the same (jingle) and calling the same thing by different names doesn't make them different (jangle)
en.wikipedia.org/wiki/Jingle-...
(this is going to be so useful for reviewing)
Des presenting at VisCon CVPR 2025
Sad not to be there in person but this work will also be presented at ACL in Vienna 2025 - see you there!
📯 Best Paper Award at CVPR workshop on Visual concepts for our (@doneata.bsky.social + @delliott.bsky.social) paper on probing vision/lang/ vision+lang models for semantic norms!
TLDR: SSL vision models (swinV2, dinoV2) are surprisingly similar to LLM & VLMs even w/o lang 👀
arxiv.org/abs/2506.03994
Paper title "Cultural Evaluations of Vision-Language Models Have a Lot to Learn from Cultural Theory"
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
Check out our new paper led by @srishtiy.bsky.social and @nolauren.bsky.social! This work brings together computer vision, cultural theory, semiotics, and visual studies to provide new tools and perspectives for the study of ~culture~ in VLMs.
as an extra take-away, this implies that our eval tends to be overly precision focused. we should really think of what we lose in terms of recalls, as this directly relates to what we miss out for whom when we build these large-scale, general-purpose models.
(4/4)
🚀 We are excited to introduce Kaleidoscope, the largest culturally-authentic exam benchmark.
📌 Most VLM benchmarks are English-centric or rely on translations—missing linguistic & cultural nuance. Kaleidoscope expands in-language multilingual 🌎 & multimodal 👀 VLMs evaluation
Join us and revolutionize Life Science Lab Automation! 🎓🤖💉
I am hiring a Postdoc in Robotics and Computer Vision for Life Science Laboratory Automation, in Copenhagen, Denmark.
Is that you? 🙋♀️
efzu.fa.em2.oraclecloud.com/hcmUI/Candid...
Yes! It's the monolithic nature of the single value system that is the target of alignment that's so problematic. (But then we also have to agree to be ok with models that generate content that we as individuals are extremely un-aligned with, right?)
Today we are releasing Kaleidoscope 🎉
A comprehensive multimodal & multilingual benchmark for VLMs! It contains real questions from exams in different languages.
🌍 20,911 questions and 18 languages
📚 14 subjects (STEM → Humanities)
📸 55% multimodal questions
The Panopticon is amazing! and thanks for this thread - my libby holds list just got a bit longer :-)
We are looking for two PhD students at our institute in Munich.
Both postions are open-topic, so anything between cognitive science and machine learning is possible.
More information: hcai-munich.com/PhDHCAI.pdf
Feel free to share broadly!
📢Excited to announce our upcoming workshop - Vision Language Models For All: Building Geo-Diverse and Culturally Aware Vision-Language Models (VLMs-4-All) @CVPR 2025!
🌐 sites.google.com/view/vlms4all
BirdCLEF25: Audio-based species identification focused on birds, amphibians, mammals, and insects in Colombia.
👉 www.kaggle.com/competitions...
@cvprconference.bsky.social @kaggle.com
#FGVC #CVPR #CVPR2025 #LifeCLEF
[1/4]
Above: Casing of Ironoquia dubia (RMNH.INS.1544419) collected on May 18th 1971 in Loenen, The Netherlands. b) The label of the specimen. Depicted on the right: detail of the artificial items. Photographs: overview: Auke-Florian Hiemstra, details: Pasquale Ciliberti. Below: Caddisfly larvae in the studio of Hubert Duprat, carrying cases made from mostly gold. © Hubert Duprat, adagp, 2024, Courtesy the Artist and Art : Concept, Paris, Photo F. Delpech.
Thanks to these insects, we can now study environmental microplastics retrospectively. 🔍 Even before Duprat began his now famous experiments with caddisfly larvae, insects in the wild were already experimenting with plastic... 🐛 14/x
Now your browser can look like Vivaldi! (except maybe the floating video thing) ❤️
What a weekend to find Heinrich von Kleist's Erzählungen next to the skip, in which the first story is literally about a man wreaking murderous havoc because of the imposition of arbitrary trade tariffs en.wikipedia.org/wiki/Michael...
Portland, ME showed up. #HandsOff
