CUDA-L2: Surpassing cuBLAS Performance for Matrix Multiplication Through RL
🥳 Introduction CUDA-L2 is a system that combines large language models (LLMs) and reinforcement learning (RL) to automatically optimize Half-precision General Matrix Multiply (HGEMM) CUDA kernels. CUDA-L2 systematically outperforms major matmul baselines to date, from the widely-used torch.matmul to state-of-the-art NVIDIA closed-source libraries (cuBLAS, cuBLASLt-heuristic, cuBLASLt-AutoTuning). Paper Speedup of CUDA-L2 over torch.matmul, cuBLAS, cuBLASLt-heuristic, and […]
Plane crashed after 3D-printed part collapsed
A plane crashed after a 3D-printed part softened and collapsed, causing its engine to lose power, a report has found. The Cozy Mk IV light aircraft was destroyed after its plastic air induction elbow, bought at an air show in North America, collapsed. The aircraft crashed into a landing aid system at Gloucestershire Airport in […]
The “confident idiot” problem: Why AI needs hard rules, not vibe checks
We cannot fix probability with more probability. We need to re-introduce “assert”. The Lie We have all been there. You build an agent. It works perfectly in the demo. You deploy it. And then, on a Tuesday at 3 PM, it decides that the URL for the API documentation is api.stripe.com/v1/users (a 404), but it […]
The RAM Shortage Comes for Us All
Memory price inflation comes for us all, and if you’re not affected yet, just wait. I was building a new PC last month using some parts I had bought earlier this year. The 64 Gigabyte T-Create DDR5 memory kit I used cost $209 then. Today? The same kit costs $650! Just in the past week, […]
I have been writing a niche history blog for 15 years
A special appeal to support Res Obscura on its Crystal Anniversary I was 25 years old when I started writing the blog version of Res Obscura, which ran from 2010 to 2023 (and still exists here). This was the early summer of 2010. I was a second-year PhD student in history, living with two roommates […]
Why are 38 percent of Stanford students saying they’re disabled?
The students at America’s elite universities are supposed to be the smartest, most promising young people in the country. And yet, shocking percentages of them are claiming academic accommodations designed for students with learning disabilities. In an article published this week in The Atlantic, education reporter Rose Horowitch lays out some shocking numbers. At Brown […]
PyTogether: Collaborative lightweight real-time Python IDE for teachers/learners
PyTogetherGoogle docs for Python. A fully browser-based collaborative Python IDE with real-time editing, chat, and visualization. pytogether.org (https://pytogether.org/) Real-time Collaboration – Edit Python code together instantly using Y.js. Secure Authentication – Log in manually or with Google OAuth. Groups & Projects – Organize your work into teams and projects. Live Drawings – Draw directly on […]
Converge (YC S23) is hiring a martech expert in NYC
Converge is building the definitive Growth OS: We help DTC Growth teams understand which marketing efforts drive profitable growth. We are the only platform combining best-in-class tracking with blended reporting and multi-touch attribution. Our unique positioning has led to rapid growth in both number and size of customers. One of the secrets of our growth […]
Multivox: Volumetric Display
This is the code I currently use to drive my volumetric displays. It supports two closely related devices which are configured in the src/driver/gadgets directory: Rotovox is a 400mm Orb featuring two 128×64 panels arranged vertically side by side. Vortex is a 300mm Orb featuring two 128×64 panels arranged horizontally, back to back. Rotovox has […]
Feynman vs. Computer
What’s neat is we can still split up the computation like we did before, if we believe it will make the error smaller and the confidence interval narrower. Let’s use the following integral as an example. [int_0^infty frac{sin{x}}{x} mathrm{d}x] This oscillates up and down quite a bit for small (x), and then decays but still […]