Lightnews — Scholar-powered news

Reposted by Emile van Krieken

Jaap Jumelet @jumelet.bsky.social · 4h

🌍Introducing BabyBabelLM: A Multilingual Benchmark of Developmentally Plausible Training Data!

LLMs learn from vastly more data than humans ever experience. BabyLM challenges this paradigm by focusing on developmentally plausible data

We extend this effort to 45 new languages!

1 10 22

Reposted by Emile van Krieken

Martin Trapp @trappmartin.bsky.social · 26d

Unfortunately, our submission to #NeurIPS didn’t go through with (5,4,4,3). But because I think it’s an excellent paper, I decided to share it anyway.

We show how to efficiently apply Bayesian learning in VLMs, improve calibration, and do active learning. Cool stuff!

📝 arxiv.org/abs/2412.06014

Post-hoc Probabilistic Vision-Language Models

Vision-language models (VLMs), such as CLIP and SigLIP, have found remarkable success in classification, retrieval, and generative tasks. For this, VLMs deterministically map images and text descripti...

arxiv.org

2 14 48

Emile van Krieken @emilevankrieken.com · 26d

Ohhh this is very relevant to what I'm currently working on! Thanks for sharing :)

1 1

Emile van Krieken @emilevankrieken.com · 26d

What's wrong with this?

1

Emile van Krieken @emilevankrieken.com · 26d

Accepted to NeurIPS! 😁

We will present Neurosymbolic Diffusion Models in San Diego 🇺🇸 and Copenhagen 🇩🇰 thanks to @euripsconf.bsky.social 🇪🇺

Emile van Krieken @emilevankrieken.com · May 21

We propose Neurosymbolic Diffusion Models! We find diffusion is especially compelling for neurosymbolic approaches, combining powerful multimodal understanding with symbolic reasoning 🚀

Read more 👇

8

Reposted by Emile van Krieken

Adam L @lauretig.bsky.social · 29d

A really neat paper thinking through what "Identifiability" means, how we can determine it, and implications for modeling.

Arxiv: arxiv.org/abs/2508.18853

#statssky #mlsky

A schematic showing some techniques for assessing identifiability and how computation-
ally expensive they are for computational models of different types, ranked in a typical order of
computational complexit

A diagram of implication conditions between identifiability concepts discussed. Note
how structural unidentifiability makes practical identifiability impossible, both locally and globally.
Conversely, practical global identifiability, that may be (loosely) tested as described in the ‘Brute
force checks . . . ’ section, would guarantee the other identifiability conditions

9 23

Reposted by Emile van Krieken

Leshem (Legend) Choshen @ICML @ACL @lchoshen.bsky.social · Sep 12

The most expensive part of training is the data, not the compute
Nikhil Kandpal & Colin Raffel calculate a really low bar for how much it would cost to produce LLM training data with 3.8$\h
Well, several scales more than the compute.
Luckily (?), companies don't pay for the data
🤖📈🧠

1 2 1

Emile van Krieken @emilevankrieken.com · Sep 11

Organising NeSy 2025 together with some of my favourite people (@e-giunchiglia.bsky.social @lgilpin.bsky.social @pascalhitzler.bsky.social) really was a dream. Super proud of what we've achieved! See you next year 😘

NeSy 2025 Conference @nesyconf.org · Sep 11

🦕NeSy 2025 is officially closed! Thanks again to everyone for attending this successful edition 😊

We will see you 1-4 September in another beautiful place: Lisbon! 🇵🇹
nesy-ai.org/conferences/...

1 10

Reposted by Emile van Krieken

NeSy 2025 Conference @nesyconf.org · Sep 6

Our second keynote speaker is @tkipf.bsky.social, who will discuss object-centric representation learning!

Do objects need a special treatment for generative AI and world models? 🤔 We will hear on Monday!

1 8

Reposted by Emile van Krieken

NeSy 2025 Conference @nesyconf.org · Sep 5

It is almost time to welcome you all in Santa Cruz! 🦕

We will start with an exciting and timely keynote by
@guyvdb.bsky.social
on "Symbolic Reasoning in the Age of Large Language Models" 👀

📆 Full conference schedule: 2025.nesyconf.org/schedule/

1 5 18

Reposted by Emile van Krieken

François Rozet @francois-rozet.bsky.social · Sep 3

Does a smaller latent space lead to worse generation in latent diffusion models? Not necessarily! We show that LDMs are extremely robust to a wide range of compression rates (10-1000x) in the context of physics emulation.

We got lost in latent space. Join us 👇

1 8 27

Reposted by Emile van Krieken

EurIPS Conference @euripsconf.bsky.social · Aug 29

We are immensely excited to announce our 4 leading researchers within our community as the #EurIPS keynote speakers🎉

@ulrikeluxburg.bsky.social

Michael Jordan

Emtiyaz Khan

Amnon Shashua

More details to come as we get closer to December, so stay tuned

7 33

Reposted by Emile van Krieken

Gergely Orosz @gergely.pragmaticengineer.com · Aug 27

Ironic: as LLMs make writing much easier, I am finding not more, but less interesting and novel things to read online.

And so I keep paying more attention to the fewer people who still write their original thoughts (without LLMs - you can tell how repetitive it gets with them)

21 23 290

Reposted by Emile van Krieken

Sophie Huiberts @sophie.huiberts.me · Aug 20

This is a good feed to have in the mixer. It shows all papers being shared by your follows

Paper Skygest Team @paper-feed.bsky.social · Aug 19

**Please repost** If you're enjoying Paper Skygest -- our personalized feed of academic content on Bluesky -- we'd appreciate you reposting this! We’ve found that the most effective way for us to reach new users and communities is through users sharing it with their network

3 15

Reposted by Emile van Krieken

Mark J. Nelson @mm-jj-nn.bsky.social · Aug 14

Free online Neuro-Symbolic AI Summer School today and tomorrow. Looks like some interesting talks!

Neuro-Symbolic AI Summer School 2025 · Luma

Tired of the same old AI ideas and looking for something new? Or wondering if there are any deep alternatives to models that hallucinate, elude human…

lu.ma

1 3 7

Reposted by Emile van Krieken

EurIPS Conference @euripsconf.bsky.social · Aug 13

We are excited to bring #EurIPS 2025 to Copenhagen in December.

Consider becoming a sponsor and support us in making this inaugural event a success! Sponsorship packages are available and can be further customized if necessary.

Reach out if you have any questions ❔
Info: eurips.cc/become-spons...

8 18

Reposted by Emile van Krieken

Christian Wolf @chriswolfvision.bsky.social · Aug 10

Thank you! Somebody (@gracekind.net ) has finally found the perfect way to answer these "its just matrix multiplication" takes.

Also, the take "there is nothing new with deep learning, neural nets were around 50y ago" is like "nothing new with humans, amino acides were around 4.4 billion y ago".

Grace @gracekind.net · Aug 10

It’s just matrix multiplication

1 4 27

Emile van Krieken @emilevankrieken.com · Aug 9

Hahaha I didn't know this!!

1

Emile van Krieken @emilevankrieken.com · Aug 8

That's a good point!

Reposted by Emile van Krieken

Marzena Karpinska @markar.bsky.social · Aug 8

GPT-5 lands first place on NoCha, our long-context book understanding benchmark.

That said, this is a tiny improvement (~1%) over o1-preview, which was released almost one year ago. Have long-context models hit a wall?

Accuracy of human readers is >97%... Long way to go!

Screenshot of benchmark with gpt-5 on top with 68.46% accuracy.

1 6 18

Emile van Krieken @emilevankrieken.com · Aug 8

These code aspects could already be handled by manually choosing that eg parentheses should be separate tokens without committing to byte-level

1 3

Emile van Krieken @emilevankrieken.com · Aug 8

Yeah, this was also my initial thought. Probably Open ai thought the weird bugs isn't worth the overhead

4

Emile van Krieken @emilevankrieken.com · Aug 8

Already gpt4 used tools right?

1 2

Emile van Krieken @emilevankrieken.com · Aug 8

Same.
Cc @edoardo-ponti.bsky.social

Emile van Krieken @emilevankrieken.com · Aug 8

I'm honestly still surprised it happens given how much research into tokenisation there is (eg byte-level).
I guess they tried alternative tokenisation and found a bad trade-off?

6 16