Lightnews — Scholar-powered news

Marco

@mcognetta.bsky.social

2.3K followers 1.2K following 640 posts

Language and keyboard stuff at Google + PhD student at Tokyo Institute of Technology. I like computers and Korean and computers-and-Korean and high school CS education. Georgia Tech → 연세대학교 → 東京工業大学. https://theoreticallygoodwithcomputers.com/

theoreticallygoodwithcomputers.com

Posts Media Videos Starter Packs

Pinned

Marco @mcognetta.bsky.social · Jan 5

A lot of you followed me due to #NLP, but I like to post about #chess (especially computer chess), #programming (especially puzzles, code golf, etc), and machine learning.

And some less technical stuff like #Korean, #Esperanto, and #trains (mostly in Japan, just due to proximity).

1 8

Reposted by Marco

Sunnie S. Y. Kim ☀️ @sunniesuhyoung.bsky.social · 6h

Our Responsible AI team at Apple is looking for spring/summer 2026 PhD research interns! Please apply at jobs.apple.com/en-us/detail... and email [email protected]. Do not send extra info (e.g., CV), just drop us a line so we can find your application in the central pool!

Machine Learning / AI Internships - Jobs - Careers at Apple

Apply for a Machine Learning / AI Internships job at Apple. Read about the role and find out if it’s right for you.

jobs.apple.com

1 10 18

Marco @mcognetta.bsky.social · 13h

This is my only cool, non-trivial one. How to dispatch on types with the same UnionAll but not necessarily the same eltype, without knowing the types ahead of time.

Like an `f` such that:

x = B{Int64}()
y = B{Float32}()
z = C{Int64}()

f(x, y) # does something
f(x, z) # fails
f(y, z) # fails

Dispatching on Types with the Same UnionAll (but You Don’t Know the Type Beforehand)

A post about efficiently dispatching on types that are in the same UnionAll, but are possibly different concrete types (for example, A{T} and A{S}) in Julia.

mcognetta.github.io

1 1

Reposted by Marco

Chris Rackauckas @chrisrackauckas.bsky.social · 2d

MIT Julia Lab: looking for postdoctoral researchers in #julialang open source software development, scientific machine learning (#SciML), and systems biological / pharmacological modeling (#QSP) for research in equation discovery for large stiff systems.

julia.mit.edu/projects/#po...

Postdoc Projects

julia.mit.edu

3 11

Marco @mcognetta.bsky.social · 2d

I think speedrunning is one of the coolest DL challenges that are immediately accessible to anyone with a GPU.

modded-nanogpt medium world record: adding value embeddings

Adding more value-embeddings to modded-nanogpt led me to an (as of yet unofficial) modded-nanogpt medium record as seen in PR#119. In this article, I present these results, and a lot more ablations of...

snimu.github.io

Marco @mcognetta.bsky.social · 2d

TBH I've just been clicking through @workspaces.bsky.social and seeing what everyone has. Idk how all these people are shelling out for the Apple Studio lol.

Marco @mcognetta.bsky.social · 2d

What monitors do you all recommend for a home office? Mostly for coding but some gaming (not a fan of curved monitors for gaming). 32-34" with a budget of ~$550 (a bit flexible).

Considering the Samsung M70D and _maybe_ the LG 32UN880-B (this is a bit out of the price range).

Reposted by Marco

thebes @vgel.me · 2d

if you're interesting in gaining a better intuition for how llms behave at inference time, you should try logitloom🌱, the open-source tool i made for exploring token trajectory trees (aka looming) on base and instruct models! more info in thread

🌱 vgel.me/logitloom
💻 github.com/vgel/logitloom

5 25 99

Reposted by Marco

Mark Riedl @markriedl.bsky.social · 2d

Hi all,

Georgia Tech is holding its 2nd Summit on Responsible Computing, AI, and Society October 28-29th in Atlanta.

rcais.github.io

Our speakers and schedule is set. It's going to be great!

If you are planning to be in Atlanta and want to join us, registration is now open on the website.

7 21

Reposted by Marco

Manoel Horta Ribeiro @manoelhortaribeiro.bsky.social · 4d

Computer Science is no longer just about building systems or proving theorems--it's about observation and experiments.

In my latest blog post, I argue it’s time we had our own "Econometrics," a discipline devoted to empirical rigor.

doomscrollingbabel.manoel.xyz/p/the-missin...

2 9 29

Reposted by Marco

Ramon Astudillo @ramon-astudillo.bsky.social · 2d

(repost welcome) The Generative Model Alignment team at IBM Research is looking for next summer interns! Two candidates for two topics

🍰Reinforcement Learning environments for LLMs

🐎Speculative and non-auto regressive generation for LLMs

interested/curious? DM or email [email protected]

1 13 19

Marco @mcognetta.bsky.social · 2d

Oh, probabilistic programming is another really great related use case.

For example, Turing.jl

The Turing Language

Turing.jl is a probabilistic programming language and Bayesian modelling framework for the Julia programming language.

turinglang.org

Reposted by Marco

Greg Durrett @gregdnlp.bsky.social · 2d

Find my students and collaborators at COLM this week!

Tuesday morning: @juand-r.bsky.social and @ramyanamuduri.bsky.social 's papers (find them if you missed it!)

Wednesday pm: @manyawadhwa.bsky.social 's EvalAgent

Thursday am: @anirudhkhatry.bsky.social 's CRUST-Bench oral spotlight + poster

5 8

Marco @mcognetta.bsky.social · 2d

Imo the best deep learning use case in Julia is physics informed neural networks, which fits in very nicely with he original motivating fields for Julia.

But I'm a bit out of the loop. I haven't used Julia seriously since ~2023 (for various reasons; I'm still a big fan).

GitHub - SciML/NeuralPDE.jl: Physics-Informed Neural Networks (PINN) Solvers of (Partial) Differential Equations for Scientific Machine Learning (SciML) accelerated simulation

Physics-Informed Neural Networks (PINN) Solvers of (Partial) Differential Equations for Scientific Machine Learning (SciML) accelerated simulation - SciML/NeuralPDE.jl

github.com

1 3

Marco @mcognetta.bsky.social · 2d

Of things to contribute to? Or use cases?

Marco @mcognetta.bsky.social · 2d

I wouldn't for anything serious, but it is a cool language.

One thing that I think it is really nice for is getting serious open-source contributions under your belt. Since the ecosystem is so much smaller, there is a lot of accessible work to be done (whereas PyTorch is very hard to break into).

1 1

Marco @mcognetta.bsky.social · 2d

I'd love to have a symposium on the current state-of-the-art in Julia + things like a hackathon, a training competition (e.g., reproducing the CIFAR speedruns), etc.

Anyone else interested in this?

#julia #julialang

GitHub - KellerJordan/cifar10-airbench: CIFAR-10 speedruns: 94% in 2.6 seconds and 96% in 27 seconds

CIFAR-10 speedruns: 94% in 2.6 seconds and 96% in 27 seconds - KellerJordan/cifar10-airbench

github.com

Marco @mcognetta.bsky.social · 2d

I wasn't confident in correctness (though I think this has gotten much much better) or the speed (especially things like multi-GPU, etc) and it was just all around not as good as e.g., PyTorch.

Ofc PyTorch has the backing of a huge company and community, so it has not a fair comparison.

Marco @mcognetta.bsky.social · 2d

I think a Language Modeling in Julia or just a general Deep Learning in Julia would be a great option.

It's been a while since I've done DL in Julia, so things may have changed. Flux (Lux) was quite nice, but the tooling and ecosystem was lacking.

#julia #julialang

Marco @mcognetta.bsky.social · 2d

JuliaCon 2026 (Mainz, Germany) Call for Minisymposia proposals just came out.

#julia #julialang

JuliaCon 2026

JuliaCon 2026, Mainz

juliacon.org

Marco @mcognetta.bsky.social · 2d

JuliaCon 2026 (Mainz, Germany) Call for Minisymposia proposals just came out.

#julia #julialang

JuliaCon 2026

JuliaCon 2026, Mainz

juliacon.org

1 2

Marco @mcognetta.bsky.social · 2d

And 3 of the 4 are so bad I'd consider them borderline academic misconduct.

Marco @mcognetta.bsky.social · 2d

The sum total length of all reviews on my AAAI paper is less than the length of any of the reviews I made for my batch.

It's really great out here y'all.

1 1

Marco @mcognetta.bsky.social · 3d

Another year, another banger

Gboard ダイヤルバージョン / Gboard Dial Version

YouTube video by Google Japan

www.youtube.com

Reposted by Marco

Maria Antoniak @mariaa.bsky.social · 3d

Here’s a #COLM2025 feed!

Pin it 📌 to follow along with the conference this week!

2 17 24

Reposted by Marco

Chris Rackauckas @chrisrackauckas.bsky.social · 4d

Claude code sucks... but it's still very useful! Check out this blog post that details how Claude is deployed to solve problems in the #Julialang #SciML open source scientific computing infrastructure.

stochasticlifestyle.com/claude-code-...

#llm #vibecoding #physics #opensource #oss

Claude Code in Scientific Computing: Experiences Maintaining Julia's SciML Infrastructure - Stochastic Lifestyle

Claude Code in Scientific Computing: Experiences Maintaining Julia’s SciML Infrastructure So it’s pretty public that for about a month now I’ve had 32 processes setup on one of the 64 core 128gb RAM s...

www.stochasticlifestyle.com

2 3 13

Reposted by Marco

Marco @mcognetta.bsky.social · 4d

I've relocated to the bay. Here's a real pic of me and my crew.

1 1 8