New weekend blogpost. Some light PTX exploration, and a simple Top-K kernel.
John Wick of CUDA kernels.
USplat4D: Uncertainty Matters in Dynamic Gaussian Splatting for Monocular 4D Reconstruction tamu-visual-ai.github.io/usplat4d/
Pay attention, people: he is a top AI researcher @GoogleDeepMind After all, attention is all you need, yet it seems to be severely lacking!
We're doing all kinds of stuff with these models that the public isn't even thinking of yet.
I shared this to many friends these days: the foundation model cycle is approaching to the end.
New paper 📜: Tiny Recursion Model (TRM) is a recursive reasoning approach with a tiny 7M parameters neural network that obtains 45% on ARC-AGI-1 and 8% on ARC-AGI-2, beating most LLMs. Blog: alexiajm.github.io/2025/09/29/tin… Code: github.com/SamsungSAILMon… Paper: arxiv.org/abs/2510.04871
Every week, another batch of robot data startups appears, selling payload dirt in a gold rush. But the market dynamics that created Scale and Surge don’t exist yet in robotics. Most naive robot trajectories are literally useless, quality is far more important than quantity.
It only took 13 years, but dark mode is finally here sebastianraschka.com/blog/2021/dl-c…
- @yacinelearning (mcgill —> unhinged deep learning tech influencer on twitter)
curious to hear other successful founder stories out of grad/drop-out from quebec, some top of mind: - @gregisenberg (mcgill --> wall street survivor /acquired) - @FDavidsonT (mcgill --> sonder /ipo) - @marcbonin18 (mcgill --> planned /series b) who else am i missing?
My colleagues @druv_pai @qu_1006 and I just gave a tutorial with a total of 12 lectures on Deep Representations of (Low-Dim) Data Distributions at IAISS 2025 in Italy. Now the slides of the 12 lectures are all shared at our open book website: ma-lab-berkeley.github.io/deep-represent… This can be…
Day 3 Coded Linear Quantisation for float to int-8 from scratch! Quantised Facebook/OPT-135M model with 8-bit linear quant Code: github.com/YuvrajSingh-mi… Linear Quantisation is a simple technique that basically creates a linear mapping between the quantised values (max and min…
Bob Carpenter, who is the main dev for @mcmc_stan, finally gets "Jax-pilled" :) statmodeling.stat.columbia.edu/2025/10/03/its… With cheap (even free) GPUs on Google Colab and Lightning AI studio, it is now easy to get 10x-100x speedups for HMC, and more if you switch to SVI.
Terence Tao + AI for solving hard math problems🤯 In this example the insight comes from Terence, and the muscle—via an hourlong conversation with GPT-5 and some python code it wrote—comes from AI. "Here, the AI tool use was a significant time saver... Indeed I would have been…
see? same pattern : - super duper expert know what the heck he’s doing - cut the project into micro functional sub tasks - feed to the llm and iterate on the context - brush up the output and ship it
As an open book, we welcome all people to contribute, to revise, and to critique. Meanwhile, as we are teaching this subject, we will also open source all teaching materials (lectures and tutorials) on the website. We encourage all colleagues to use and to contribute to its…
We wrote a book about representation learning! It’s fully open source, available and readable online, and covers everything from theoretical foundations to practical algorithms. 👷♂️ We’re hard at work updating the content for v2.0, and would love your feedback and contributions
it's insane to me how little attention the llm.q repo has it's a fully C/C++/CUDA implementation of multi-gpu (zero + fsdp), quantized LLM training with support for selective AC it's genuinely the coolest OSS thing I've seen this year (what's crazier is 1 person wrote it!)
episode 1: drop resnet lore on arxiv is schmidhuber now jacked, smart, AND funny?
i'm really looking forward to the schmidhuber revenge arc from the sands of arabia
Yet more evidence that a pretty major shift is happening, this time by Scott Aaronson scottaaronson.blog/?p=9183&fbclid…
Douglas Hofstadter wrote about his experience of running up against an “abstraction ceiling” in his own brain while pursuing a PhD in mathematics. As Hofstadter describes, the abstraction ceiling is not a “hard” threshold, a level at which one is suddenly incapable of learning…
Your mathematical potential has a limit, but it's likely higher than you think. Most people can learn basic math like arithmetic and some algebra – but beyond that, higher levels of math become increasingly abstract and technical, and fewer people have the cognitive resources to…
Cambridge "Statistics" lecture notes PDF: statslab.cam.ac.uk/~rrw1/stats/Sa… Example sheets & more: statslab.cam.ac.uk/~rrw1/stats/in…
United States Trends
- 1. #WWERaw 61K posts
- 2. Packers 45K posts
- 3. John Cena 62.2K posts
- 4. Eagles 70K posts
- 5. Jalen 15.4K posts
- 6. #GoPackGo 4,762 posts
- 7. #RawOnNetflix 1,549 posts
- 8. Grand Slam Champion 17K posts
- 9. Jordan Love 3,794 posts
- 10. Matt LaFleur N/A
- 11. Green Bay 11K posts
- 12. Rusev 2,571 posts
- 13. Cade Horton 1,196 posts
- 14. Tush Push 10.6K posts
- 15. Kevin Patullo N/A
- 16. Shipley N/A
- 17. #MondayNightFootball N/A
- 18. Dirty Dom 1,628 posts
- 19. Drake Baldwin 5,536 posts
- 20. Go Birds 17.7K posts
Something went wrong.
Something went wrong.