Mirko Klukas

Animation from my new preprint: "Fragmented Spatial Maps: State Abstraction and Efficient Planning from Surprisal" biorxiv.org/content/10.110… L: Agent traversing a hairpin maze. R: The corresponding internal representation. Red dots and lines represent remapping events.

I made something. It's like a web view, but native. Rather than parsing HTML and running JavaScript, it fetches machine code binaries over the internet and runs them sandboxed. Here's the first app: a "Top" app that runs over SSH.

@rohanpaul_ai Complete misunderstanding of what mathematics is about. If at all mathematical education becomes MORE important. Mathematics is about studying structure and learning to think about abstract ideas.

Peter Thiel on who is most likely to lose jobs to AI: "It seems much worse for the math people than the word people" "Within 3-5 years AI will be able to solve all the US Math Olympiad problems. That would shift things quite a bit."

@mrcslws Braids! en.wikipedia.org/wiki/Braid_gro…

1D manifolds.

Amazon’s Jeff Bezos has more money than could be spent in a million lifetimes. Why then does his company force workers to meet impossible quotas and treat them as disposable when they are injured? To Amazon, long-term pain and disability is just the cost of doing business.

We will have a handbook for SLAM! Together with amazing contributors, we are completing the book. The first part of the book is now available to the public. github.com/SLAM-Handbook-…

Mechazilla has caught the Super Heavy booster!

@NobelPrize ???

BREAKING NEWS The Royal Swedish Academy of Sciences has decided to award the 2024 #NobelPrize in Physics to John J. Hopfield and Geoffrey E. Hinton “for foundational discoveries and inventions that enable machine learning with artificial neural networks.”

@Ethan_smith_20 @jeremyphoward youtube.com/watch?v=i9GiLB…

@Ethan_smith_20 @Ethan_smith_20 @jeremyphoward You can use far fewer (and random) frequencies to create unique positional codes using periodical patterns. If you want to go off on a tangent look into "grid cells" in the entorhinal cortex of rats or other mammals. Happy to provide more context

i'm not sure why I haven't noticed this until now, is it not an issue that frequencies in sinusoidal positional embeddings get basically clipped passed a certain dimension? should we be using slower changing frequencies in scaling up to larger dimensions?

What's the greatest "conversation" scene from a movie?

How I fell out of love with academia (this video was an accidental publication/scheduling blunder😬😬 but well uh, happy Friday I guess) youtube.com/watch?v=LKiBlG…

I call this shot “The Birth of Vesta”. It appears in Scavengers Reign’s finale. Hands down the most insane shot I have ever animated. 😵‍💫 Color @benjybrooke @SeanBuckelew Compositing @charleshuettner ✨ Process below 🧵

Objects called contact structures provide insights into the manifolds they live on, as well as links to physics. A 1989 paper by Yakov Eliashberg answered many questions about three-dimensional manifolds, but higher dimensions remain the ‘wild west.’ quantamagazine.org/in-the-wild-we…

In 2019, the mathematicians Agustin Moreno, Fabio Gironella and Jonathan Bowden (left to right) found a wide variety of manifolds that illustrate unique relationships between contact geometry and symplectic geometry. quantamagazine.org/in-the-wild-we…

New blog post: "What happens when you vectorize wide PyTorch expressions?" probablymarcus.com/blocks/2023/10…

3D understanding is key in vision & robotics. Moving beyond common discriminative models, we introduce 3DNEL, a generative probabilistic inverse graphics take on 6D pose estimation. Result? SOTA & added probabilistic benefits. Explore at probcomp.github.io/nel & #ICCV2023 🧵

The level of incompetence in providing a smooth digital experience in Germany is simply mind-blowing. Why is this so challenging? It's 2023, I just don't get it.

KRILLERTIME ALL THE SINGLE HADES kriller.com

It's 2023. Just make publications a single decently sized column. Why would anyone want to read a two column paper where you constantly read downwards, then jump up and read downwards again, and then turn over the page to repeat that.

@Michael_J_Black Regarding trust, there should be a cultural change being more critical of overselling results and creative storytelling. Public and media are just picking up those stories. Truth is we don't know how far we are from such a model making an "actual" contribution. What am I missing?

With LLMs for science out there (#Galactica) we need new ethics rules for scientific publication. Existing rules regarding plagiarism, fraud, and authorship need to be rethought for LLMs to safeguard public trust in science. Long thread about trust, peer review, & LLMs. (1/23)