ah yes, the gen z way of going through papers: minecraft parkour in the background and the peter griffin voiceover

overleaf has probably got the worst AI (I assume) autocompletion system possible. it just keeps suggesting absolute rubbish

https://arxiv.org/abs/2503.08146

the proceedings paper for our work is available on arxiv and is also gonna be presented at CHEP I suppose

I've been thinking (and reading) a lot about visual reasoning lately. found this stream today with a nice set of papers to go through. not the latest breaking news but good enough to solidify some of my thoughts. he also talks about this paper there that's not related to visual reasoning directly but it's a nice idea of kind of like getting training data out of the model itself (he also mentions an information theoretical argument for why this is absurd in theory but I don't think it can be applied to this case)
#edu

key takeaway: it's not that quantum mechanics is confusing it's that probability theory is confusing and also likely wrong

I use neovim btw👍

Ого, біологічний кубіт...

https://www.nature.com/articles/s41586-025-09417-w

цікавий спосіб візуалізації ембедингів

https://rti.github.io/gptvis/

more info + contact your MEPs in 3-4 clicks.
(you don't need to be an EU-citizen)

september 12 is an important deadline

https://fightchatcontrol.eu/#contact-tool

man I love environmental story telling

"...Probing with sparse autoencoders on Llama 70B revealed a counterintuitive gating mechanism: suppressing deception-related features dramatically increased consciousness reports, while amplifying them nearly eliminated them..."

AI lobotomy is real