diack.bsky.social - Profile | ThreadSky | a Reddit-style client for Bluesky

The US has denied the visas of all 12 members of the Senegalese women’s basketball team who were scheduled to travel to the country as part of their preparation for the Afrobasket tournament that begins in August. Afrobasket is also the African qualifiers for the Olympics. #wnba

submitted 3 days ago • 249 comments

Towards Automating Long-Horizon Algorithm Engineering for Hard Optimization Problems Blog: sakana.ai/ale-bench/ Paper: arxiv.org/abs/2506.09050 ALE-Bench is a coding benchmark focused on NP-hard optimization problems. ALE-Agent, created using ALE-Bench, ranked 21/1000 on AtCoder Heuristic Contest!

submitted 6 days ago • 2 comments

Congratulations Andrew Rabinovich (PhD ‘08) on winning the Longuet-Higgins Prize at #CVPR2025! (1/2)

submitted 9 days ago • 2 comments

#CVPR2025 paper awards

submitted 9 days ago • 3 comments

Today we released Institutional Books 1.0, a 242B token dataset from Harvard Library's collections, refined for accuracy and usability. 🧵

submitted 10 days ago • 10 comments

🤘🏿

submitted 11 days ago • 0 comments

📢BecomingLit: Relightable Gaussian Avatars with Hybrid Neural Shading📢 We propose a hybrid neural shading scheme for creating intrinsically decomposed 3DGS head avatars, that allow real-time relighting and animation. 🌍https://lnkd.in/evNt8bV2 📷https://lnkd.in/ekB5QeEK

submitted 13 days ago • 1 comment

Canada is experiencing multiple wildfires www.google.com/maps/@54.111... Research blog on realtime wildfire tracking: research.google/blog/real-ti...

submitted 18 days ago • 1 comment

Hello #MLSky! A new class of generative models for you! Check out Alternators 🌀 They outperform Mambas and diffusion models on time-dependent modeling tasks. I am excited about all the possibilities with this new class of models. #Alternators #GenAI

submitted 18 days ago • 0 comments

Announcing the release of the official Common Corpus paper: a 20 page report detailing how we collected, processed and published 2 trillion tokens of reusable data for LLM pretraining arxiv.org/pdf/2506.01732

submitted 18 days ago • 3 comments

Why human connectomes can probably not be useful for AI open.substack.com/pub/kording/...

submitted 18 days ago • 8 comments

“Vibe debt”

submitted 19 days ago • 17 comments

🤯 We use the term 'intelligence' a lot, but wth do we mean? We got 303 survey responses from researchers. The most agreed-on criteria are generalization, adaptability & reasoning. ACL Findings preprint: arxiv.org/abs/2505.20959 with @brtrm.bsky.social @terne.bsky.social @heinrichst.bsky.social /1

submitted 20 days ago • 3 comments

Fascinating Ghana COVID study! 🇬🇭 80% of participants had prior infection, leading to STRONGER & more durable antibody responses post-vaccine vs UK! Vaccines worked & boosted immunity. Intriguing finding: higher income linked to more COVID. bmcmedicine.biomedcentral.com/articles/10....

submitted 22 days ago • 0 comments

2023: Gaussian splatting paper is published, triangles are so over … 2025: Triangles are so back babbbyyyy trianglesplatting.github.io

submitted 23 days ago • 1 comment

I find it amusing when AI customer service agents make fake typing sounds

submitted 25 days ago • 0 comments

Meet another one of our graduating undergrads, Aminah Aliu🐯🧡 For her thesis, Aminah leveraged the Vendi Score & the Vendiscope to study embedding models on Yoruba datasets. She was 1 of four Princeton students nominated for the CRA Undergraduate Research Award and won an honorable mention. 1/2

submitted 26 days ago • 1 comment

🚀🚀🚀Announcing our $13M funding round to build the next generation of AI: 𝐒𝐩𝐚𝐭𝐢𝐚𝐥 𝐅𝐨𝐮𝐧𝐝𝐚𝐭𝐢𝐨𝐧 𝐌𝐨𝐝𝐞𝐥𝐬 that can generate entire 3D environments anchored in space & time. 🚀🚀🚀 Interested? Join our world-class team: 🌍 spaitial.ai youtu.be/FiGX82RUz8U

submitted 26 days ago • 4 comments

AI flood forecasts for Romania (next few days): sites.research.google/floods/l/44....

submitted 26 days ago • 0 comments

Wow - french Ministry of Culture releases a big dataset of their Compare:IA french llm arena tests aimed at getting good french LLMs huggingface.co/datasets/min...

submitted 29 days ago • 0 comments

Offline, small, multimodal AI on devices has real potential. I'm keen to see the practical apps people will build, especially in my region. developers.googleblog.com/en/introduci...

submitted 32 days ago • 0 comments

🚨Pls share! Join us next Thurs (22nd May 5.20pm UK time) to watch a panel discussion on our NeuroArchitecture Project: 100 Minds in Motion 100+ people explored, navigated & evacuated an art gallery kitted out with a range of sensors First findings revealed Register: www.ucl.ac.uk/bartlett/res...

submitted 37 days ago • 3 comments

Jules.google

submitted 33 days ago • 0 comments

Another example of opportunistic A.I. applied to medical imaging. Quantifying coronary artery calcium score from chest CT (nongated, not performed for this purpose) ai.nejm.org/doi/full/10....

submitted 37 days ago • 1 comment

Meanwhile in Europe...

submitted 37 days ago • 0 comments

😎

submitted 37 days ago • 0 comments

In vivo gene editing 😲 #MedSky www.nejm.org/doi/full/10....

submitted 38 days ago • 2 comments

#NewPaper We revisit Shannon's mutual information, highlight its shortcomings, and propose a novel alternative to it, called the Vendi Information Gain (VIG), using the Vendi Score. #InformationTheory #Diversity By: Quan Nguyen & @adjiboussodieng.bsky.social Link to paper: arxiv.org/abs/2505.09007

submitted 38 days ago • 0 comments

My path into AI The sort of small wins that accumulate into a real career in AI. When I started grad school AI prof's didn't have space for me in their group and when I ended I had no papers at NeurIPS/ICLR/ICML, yet the process can still work. www.interconnects.ai/p/my-path-in...

submitted 39 days ago • 1 comment

Announcing AlphaEvolve, our new LLM coding agent that has - made new scientific discoveries - discovered algorithms that are now deployed at Google (in Gemini, Transformers, TPU hardware design & data centers) Blog: deepmind.google/discover/blo... White paper: storage.googleapis.com/deepmind-med...

submitted 39 days ago • 5 comments

More updates to my WebGPU Particle Life simulator! ✅ Optional central potential ✅ Optional symmetric forces (that don't violate Newton's laws) ✅ Visual enhancements (yep, I've settled on the shiny variant) ✅ FPS throttling to prevent latency & hangs lisyarus.github.io/webgpu/parti...

submitted 43 days ago • 1 comment

🎉 Thrilled to share our CVPR 2025 Award Candidate & Oral paper: 🔹 GlobustVP Convex Relaxation for Robust Vanishing Point Estimation in Manhattan World 🧱 Global optimality 💥 Tolerates up to 70% outliers ⚡ Fast runtime 📄 Paper: arxiv.org/abs/2505.04788 💻 Code: github.com/WU-CVGL/GlobustVP 1/

submitted 44 days ago • 1 comment

what do you teach at the first-year graduate course on machine learning, in this era of LLM and large-scale compute? here's my experiment on answering this question: let's teach everything that admits SGD and that is not LLM, and ask students to read old papers.

submitted 45 days ago • 1 comment

#ICML2025 Is standard RLHF optimal in view of test-time scaling? Unsurprisingly no. We show a simple change to standard RLHF framework that involves 𝐫𝐞𝐰𝐚𝐫𝐝 𝐜𝐚𝐥𝐢𝐛𝐫𝐚𝐭𝐢𝐨𝐧 and 𝐫𝐞𝐰𝐚𝐫𝐝 𝐭𝐫𝐚𝐧𝐬𝐟𝐨𝐫𝐦𝐚𝐭𝐢𝐨𝐧 (suited to test-time procedure) is optimal!

submitted 45 days ago • 1 comment

Kenya’s high court recently ruled that Worldcoin, the biometric startup co-founded by Sam Altman, has to delete the personal data of over 300,000 Kenyans. Learn more about how Worldcoin operated in Kenya and other developing countries in this story from 2022. ⬇️⬇️⬇️

submitted 45 days ago • 1 comment

submitted 45 days ago • 0 comments

🚨New Preprint! Did you know that steering vectors from one LM can be transferred and re-used in another LM? We argue this is because token embeddings across LMs share many “global” and “local” geometric similarities!

submitted 46 days ago • 3 comments

Ah, all the artifacts were fixed by replacing one '<' with '<=', duh. It all works now, particle life completely on WebGPU! There are 2^18 (~260k) particles here; my PC struggles after about half a million particles. Will probably try to optimize it a bit later.

submitted 47 days ago • 4 comments

Today we’re sharing an early look at our latest Gemini model update for I/O! Introducing the updated Gemini 2.5 Pro (I/O edition), which ranks #1 on WebDev Arena and surpasses our previous 2.5 Pro model by +147 Elo points. 🏆 blog.google/products/gem...

submitted 47 days ago • 6 comments

LLMs are enough of an improvement to my workflows that i find people who call them useless really funny. Yes they make mistakes, but so do i, and the loop to fixing those mistakes is shorter with their help.

submitted 48 days ago • 43 comments

Generative AI has been used to design diverse mitochondrial targeting sequences, expanding options for metabolic engineering, protein delivery, and studies of mitochondrial function. doi.org/g9hf74

submitted 48 days ago • 0 comments

📢 New paper CVPR 25! Can meshes capture fuzzy geometry? Volumetric Surfaces uses adaptive textured shells to model hair, fur without the splatting / volume overhead. It’s fast, looks great, and runs in real time even on budget phones. 🔗 autonomousvision.github.io/volsurfs/ 📄 arxiv.org/pdf/2409.02482

submitted 48 days ago • 1 comment

🎶 365 jours puis 700, le temps commence à manquer, tu l'espionnes 🎶

submitted 48 days ago • 0 comments

Through vibe coding, I'm getting a sense of how good Three.js is for visualization. Fascinating to see the creative ways Gemini generated this visualization for Masked Autoencoder concepts – such an epic paper btw.

submitted 49 days ago • 1 comment

Magnitude : 7.5 Region: Drake Passage Time: 2025-05-02 12:58:26 UTC Epicenter : 68.18°W 56.85°S Depth: 10 km First posted at: 13:05 UTC https://geofon.gfz-potsdam.de/eqexplorer/events/gfz2025iobo/general #GEOFON #GFZ #earthquake #magnitude

submitted 51 days ago • 0 comments

Brazil!! 🎉

submitted 55 days ago • 0 comments

Check out our new paper at #ICLR2025, where we show that multi-task neural decoding is both possible and beneficial. As well, the latents of a model trained only on neural activity capture information about brain regions and cell-types. Step-by-step, we're gonna scale up folks! 🧠📈 🧪 #NeuroAI

submitted 58 days ago • 0 comments