Hosung Lee

MS student @ KAIST MLML

[Elliot Arledge @elliotarledge]

Introducing Netherite!

Minecraft 1.11.2, rewritten from scratch in C and CUDA, bit-verified against the real game.

This is one trained agent playing in it - then the renderer it actually trains through - then 7,200 live worlds stepping in lockstep on one GPU.

[Lean @leanprover]

Lean 4.32.0 is out with 102 changes, including:

The new do elaborator, which has been opt-in since 4.29, is now the default. The legacy elaborator is still available via https://t.co/3TJxwqIVmY.legacy if you need it.

A new module linter framework (checks that run once per module instead of after every command), a round of fixes to mvcgen' and grind for combining verification and proof automation, and a fix that cuts Mathlib import time by roughly 10%.

Full release notes: https://t.co/1jTAon02pH

#LeanLang #LeanProver #FormalVerification

[Yisong Yue @yisongyue]

Synchronization is gaining traction as an interesting way to shape token dynamics through attention layers.

Check out this great work at #ICML2026 🇰🇷

🗓️ This Thursday, 10:45 AM KST, Hall A

[FFmpeg @FFmpeg]

FFmpeg's native AAC encoder has just been rewritten, and now beats both fdkaac and qAAC according to current metrics and listening tests.
This is not a small change. @X and @OBSProject use it, as well as many others. It's been a critical piece of the internet, and is now the best

[Alex Nichol @unixpickle]

One of their claims is that their model is very parameter efficient, but they don't actually do an apples-to-apples comparison to an equally small conventional NN. I ran a small experiment on MNIST, and an MLP is actually the same or better than coupled oscillators.

[Stéphane Deny @StphTphsn1]

"Neural Fields as World Models"
by Joshua Nunley (Feb 2026)

🔗https://t.co/KMnrwRk8OM

[Sungjin Ahn @SungjinAhn_]

🚀 We introduce Neural Theorizer (NEO) — a new type of world model that learns to theorize the world from observation, without language or LLM supervision.

Selected as an ICML 2026 oral presentation — 0.7% of submitted papers.

The paper asks:

"What does it mean to understand the world and build a world model?"

Today’s world models are often trained to predict the future: the next frame, next latent state, or next observation.

But is prediction enough?

We argue that a world model should be a theory-building system: one that discovers reusable primitives, composes them into executable explanations, and transfers those explanations to novel phenomena.

NEO is our first step toward this vision — a World Theory Model that learns explicit, compositional theories from raw observation.

This work was led by my wonderful students: Doojin Baek*(@doojin_a_baek), Gyubin Lee* (@gyubin0521), Junyeob Baek (@JunyeobB), and Hosung Lee (@HosungLee_).

For more details, take a look at the paper — and if you’re attending ICML, let’s talk there!

📄 arXiv: https://t.co/TGMXLLfzP7
🌐 Project page: https://t.co/aLJywp8rfq