Is there any activity here whatsoever? *pinging for signs of life*

Lecture slides for my "Introduction to #ComputerVision" and "#DeepLearning in Computer Vision" courses. 🆕 Gaussian Splatting 🆕 Flow Matching The included videos do not contain voiceovers yet, planned for a future revision.

Relaxing Munich

"What you don't know may not hurt you, but what you don't remember always does."

Swagger supports OpenAPI 3.1 since 19 Jun 2023 (at least that's the blog post date announcing it). You think you can use the latest Swagger Editor for OpenAPI 3.1 on 22 April 2025, only 673 days later? Nope, only if you pull tag `next-v5-unprivileged` ¿¡

1. LLM-generated code tries to run code from online software packages. Which is normal but 2. The packages don’t exist. Which would normally cause an error but 3. Nefarious people have made malware under the package names that LLMs make up most often. So 4. Now the LLM code points to malware.

🦀 Help us create a vision for Rust's future, by taking part in our vision survey! Check out the blog post: blog.rust-lang.org/2025/04/04/v...

Plan wisely.

This looks like a paper worth diving in "We __prove__ that GNNs, trained to minimize a sparsity-regularized loss over a small set of shortest path instances, exactly implement the Bellman-Ford (BF) algorithm for shortest paths."

Let's go even further in the past to 2005 with David Silver's "Cooperative Pathfinding" paper Yes, the same David Silver at DeepMind who led many of the breakthrough papers including AlphaGo and AlphaZero

Why only mention new papers when posting? Here's CBS, cited *check notes* 1521 times as of today. "Conflict-based search for optimal multi-agent pathfinding"

🦀 #rustlang 1.86.0 has been released! ✨ This one has a few very nice new things, including trait upcasting, Vec::pop_if, and get_disjoint_mut! blog.rust-lang.org/2025/04/03/

New post: The Fifth Kind of Optimisation tratt.net/laurie/blog/...

Jetson @nvidia's version of our robot is available! Compute is now on-board like a @Tesla car with FSD 🚗 Importantly, we rethink the control interface, so that you can view the video stream with the wonderful @rerundotio ⭐

PEP 751 has been accepted! peps.python.org/pep-0751/ This means #Python now has a lock file standard that can act as an export target for tools that can create some sort of lock file. And for some tools the format can act as their primary lock file format as well instead of some proprietary format.

New paper for the day "RAILGUN: A Unified Convolutional Policy for Multi-Agent Path Finding Across Different Environments and Tasks" Interesting part is: "RAILGUN is not an agent-based policy but a map-based policy."

[1/10] Is scene understanding solved? Models today can label pixels and detect objects with high accuracy. But does that mean they truly understand scenes? Super excited to share our new paper and a new task in computer vision: Visual Jenga! 📄 arxiv.org/abs/2503.21770 🔗 visualjenga.github.io

Let's see if posting here about Multi-Agent Systems would get more interactions than Twitter. Starting with a review paper I wrote together with Keisuke Okumura "A Comprehensive Review on Leveraging Machine Learning for Mutli-Agent Path Finding", published in IEEE Access.

Ants are impressive in swarms.

New paper! We merge SfM reconstructions with point cloud registration. Link: arxiv.org/abs/2503.17093 Code: Not yet public, but coming later.

I believe the closest to a video of a wizard we will ever see is the Numberphile series with Neil Sloane (www.youtube.com/playlist?lis...). That's how I believe wizards behave. (N.S. who, btw, did his PhD on Neural Networks before all of us were born 🤯 neilsloane.com/doc/a000435_...)

On the other hand... bsky.app/profile/chri...

You can't effectively plan or act in the world if you can't accurately perceive it. Perception precedes planning and action. FoundationStereo looks awesome! (Screenshot from arXiv:2404.12440)

There's an obvious race to be first to successfully prop up an LLM as the source of truth and then use it to serve propaganda

It turns out that = trust me bro

This is a reminder for you to just ship your side projects. It can't be worse than such an Apple decision. (Although I believe it's all marketing stunts for shareholders to see that their money is going into AI somehow, but come on Apple...)

Code and data are now online for CameraHMR, our state-of-the-art parametric 3D human pose and shape (HPS) estimation method that will appear at hashtag#3DV2025. github.com/pixelite1201...

Finally got my copy! “AI Engineering” is officially out 🙏 🎉 It’s heavier than I expected (500 pages) and I’m so glad O’Reilly decided to publish it in color. Thanks everyone for making this happen! Thank you for giving this book a chance!

Turns out we weren't done for major LLM releases in 2024 after all... Alibaba's Qwen just released QvQ, a "visual reasoning model" - the same chain-of-thought trick as OpenAI's o1 applied to running a prompt against an image Trying it out is a lot of fun: simonwillison.net/2024/Dec/24/...

What if we put a pool on the moon? Watch the latest What If? video collaboration with Minute Earth! www.youtube.com/watch?v=aIIB...

If you take one advice for more focused work in 2025 let it be this: Take breaks that are more boring than the task at hand. Your goal is to make the task release more dopamine than the break. Even if your break means staring at a wall. Your brain will thank you.

Introducing ASAL: Automating the Search for Artificial Life with Foundation Models Blog: sakana.ai/asal/ We propose a new method called Automated Search for Artificial Life (ASAL) which uses foundation models to automate the discovery of the most interesting and open-ended artificial lifeforms!

There's nothing like the feeling of starting a codebase from scratch.

As hard as it is to believe, this dog is the OG booster of open source AI.

Six months ago someone put a for-loop around GPT-4o and got 50% on the ARC-AGI test set and 72% on a held-out training set redwoodresearch.substack.com/p/getting-50... Just sample 8000 times with beam search. o3 is probably a more principled search technique...

A happy author discovering the first hard copies

Making PRs like this is so satisfying 🔥😎

This is such an interesting angle, because the number of large organizations that have this level of discipline to evaluating new models is likely to be tiny Running a bunch of test prompts specific to what your company does through a new model feels like it should be pretty low hanging fruit

“There is nothing quite so useless, as doing with great efficiency, something that should not be done at all.”

I'm looking for an intern! If you are: * Driven * Love OSS * Interested in distributed PyTorch training/FSDPv2/DeepSpeed Come work with me! Fully remote, more details to apply in the comments

Small yet mighty! 💫 We are releasing SmolVLM: a new 2B small vision language made for on-device use, fine-tunable on consumer GPU, immensely memory efficient 🤠 We release three checkpoints under Apache 2.0: SmolVLM-Instruct, SmolVLM-Synthetic and SmolVLM-Base huggingface.co/collections/...

Welcome to German Bureaucracy: Sick with fever, it's 3°C outside. Had to cancel Physiotherapy appointment. Physio people call, they need a paper from my general doctor. Call general doctor, they need me to be present with insurance card. Second option? 👇🏻