Simon Pepin Lehalleur @plain_simon

Mathematician (algebraic geometry, motives & friends, singularities in statistics and ML). 'Geometry is successful magic' (R. Thom) University of Amsterdam. simon-pepin.github.io Nijmegen, Netherlands Joined December 2019

Tweets

16K
Followers

4K
Following

6K
Likes

125K

Keenan Crane @keenanisalive

9 hours ago

“Everyone knows” what an autoencoder is… but there's an important complementary picture missing from most introductory material. In short: we emphasize how autoencoders are implemented—but not always what they represent (and some of the implications of that representation).🧵

17 171 1K 75K 1K

Download Image

Simon Pepin Lehalleur @plain_simon

12 hours ago

Acyn @Acyn

a day ago

https://t.co/54rw7Y9nH4

4K 3K 28K 21.8M 6K

Download Video

0 0 3 320 0

Download Gif

François Rozet @FrancoisRozet

4 days ago

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 👇

14 101 455 86K 338

Download Image

arXiv math.AG Algebraic Geometry @mathAGb

3 days ago

Tom Biesbrouck, Nero Budur, Johannes Nicaise, Willem Veys: Birational zeta functions arxiv.org/abs/2509.03352 arxiv.org/pdf/2509.03352 arxiv.org/html/2509.03352

0 2 17 781 2

Boaz Barak @boazbaraktcs

3 days ago

Giving first lecture in my AI safety course tomorrow. I plan to record and post lectures online on the course webpage boazbk.github.io/mltheorysemina…

15 66 621 49K 537

🦋/acc 🌲 @argyros_selini

3 days ago

Sam Altman @sama

3 days ago

https://t.co/aNpYlXwYIa

4K 2K 35K 5.6M 3K

28 1K 23K 511K 2K

Download Image

Ulkar @ulkar_aghayeva

6 months ago

the fact that so many domains are amenable to spatial metaphors (fitness landscape, Waddington’s epigenetic landscape, decision trees, graphs, phase space, latent space) probably has nothing to do with the role of hippocampus in navigation and memory

4 2 60 13K 26

Jules Jacobs @JulesJacobs5

4 days ago

When generating tree data structures for randomized testing, use Boltzmann sampling with a decay hack to get nice and easily controllable trees

4 10 137 7K 99

Download Image

Daniel Litt @littmath

4 days ago

While it’s true one can elicit poor performance on basic math question from frontier models like GPT-5, IMO this kind of thing (in NYTimes) is likely to mislead readers about their math capabilities.

33 17 415 102K 44

Download Image

Surya Ganguli @SuryaGanguli

4 days ago

Varadhan’s work in large deviation theory is beautiful. If you are physics rather than pure math inclined, I also highly recommend Touchette’s review article on large deviation theory from a stat mech perspective: arxiv.org/abs/0804.0327

Probability and Statistics @probnstat

4 days ago

0 14 122 31K 32

Download Image

4 19 162 11K 104

Daniel Filan @dfrsrchtwts

4 days ago

🚨: if you apply to MATS, there is no guarantee that you won't be transformed into a 🌈lovely colourful geometric figure🌈. Behold the past fate of mentees of @danielmurfet, @jesse_hoogland, @RiechersPaul, and @adamimos. Apply to their streams, and the same fate could befall you!

2 8 103 6K 37

Download Image

Greg Egan @gregeganSF

6 days ago

If you pick two points at random inside a regular dodecahedron and draw the line that contains them, that line can intersect three different kinds of pairs of faces, separated by angles of: π, arcos(–1/√5), or arcos(1/√5)

1 8 33 4K 8

Download Gif

Luca Ambrogioni @LucaAmb

6 days ago

1/2) I am very happy to finally share something I have been working on and off for the past year: "The Information Dynamics of Generative Diffusion" This paper connects the entropy production, divergence of vector fields and spontaneous symmetry breaking in a unified framework

11 111 914 59K 748

Download Image

Jack Clark @jackclarkSF

6 days ago

Five years ago the frontier of LLM math/science capabilities was 3 digit multiplication for GPT-3. Now, frontier LLM math/science capabilities are evaluated through condensed matter physics questions. Anyone who thinks AI is slowing down is fatally miscalibrated.

81 240 2K 366K 512

Download Image

sky @skydotcs

7 days ago

the time picker on iphones alarm app isn’t actually circular it’s just a really long list

2K 6K 214K 40.8M 16K

Download Video

Daniel Litt @littmath

7 days ago

>took French in high school with the expectation that I would read great works of French literature >expectation comes true: >Bourbaki, SGA, EGA, etc.

23 9 395 16K 33

Kendric Tonn @kendrictonn

a week ago

Wake up, they invented p-zombies for sandwiches

11 21 296 12K 3

SIAM Activity Group on Dynamical Systems @DynamicsSIAM

a week ago

Lecture notes: "Modern aspects of Markov chains: entropy, curvature and the cutoff phenomenon" (by Justin Salez): arxiv.org/abs/2508.21055

2 35 199 8K 179

arXiv math.AG Algebraic Geometry @mathAGb

a week ago

Tudor P\u{a}durariu, Yukinobu Toda: The Dolbeault geometric Langlands conjecture via limit categories arxiv.org/abs/2508.19624 arxiv.org/pdf/2508.19624 arxiv.org/html/2508.19624

0 12 33 10K 2

Goodfire @GoodfireAI

2 weeks ago

Arc Institute trained their foundation model Evo 2 on DNA from all domains of life. What has it learned about the natural world? Our new research finds that it represents the tree of life, spanning thousands of species, as a curved manifold in its neuronal activations. (1/8)