Arav Kumar

on the atlas — 64

• Dust: Pretraining Transformers Without Backpropagation1 savers
• How things win — Steph Ango3 savers
• Nat Friedman60 savers
• What just happened? A retrospective of AI alignment — LessWrong4 savers
• The algorithm will see you now - Works in Progress Magazine4 savers
• AI keeps stubbornly refusing to take our jobs1 savers
• Do One New Thing A Day To Solve Your Problems - Algon332 savers
• Stop Saying Half of 2026 US Datacenter Capacity Is Canceled1 savers
• Inside vLLM: Anatomy of a High-Throughput LLM Inference System - Aleksa Gordić11 savers
• Without specific countermeasures, the easiest path to transformative AI likely leads to AI takeover - LessWrong7 savers
• Mark Zuckerberg - Colossus14 savers
• My Model Of EA Burnout — LessWrong3 savers
• Augmented Lagrangian Predictive Coding: training 1000-layer networks without backpropagation1 savers
• AdaWorld4 savers
• An Advanced System Architecture Breakdown of OpenAI’s Jalapeno Inference Accelerator2 savers
• Retrospectively Reverse-Engineering Apple's Neural Engine | Eileen Yoon2 savers
• Spacecraft electric propulsion - Wikipedia2 savers
• Plasma acceleration1 savers
• Antimatter is the best post-chemical rocket propulsion system – Casey Handmer's blog1 savers
• Antimatter Development Program – Casey Handmer's blog5 savers
• On Navier–Stokes | OpenAI11 savers
• Industrial Overcapacity Enables Scientific Discovery1 savers
• Open Catalyst1 savers
• The end of the age of heroes - by Noah Smith - Noahpinion2 savers
• There has to be a better way to make titanium | Orca Notes2 savers
• A retrospective of AI alignment15 savers
• AlphaExploitem: Going Beyond the Nash Equilibrium in Poker by Learning to Exploit Suboptimal Play1 savers
• LSI Logic1 savers
• What if our biggest problems aren’t economic?1 savers
• https://power2026.ai/8 savers
• Unlearning Data at Scale2 savers
• The cuda-oxide Book — cuda-oxide1 savers
• Pavlov's List: A List of RL Environment Startups7 savers
• start [Silicon Prawn]1 savers
• The end of the age of heroes - by Noah Smith - Noahpinion1 savers
• TPU Deep Dive10 savers
• Om Gokhale3 savers
• [2402.14730] Clifford-Steerable Convolutional Neural Networks1 savers
• [2406.08391] Large Language Models Must Be Taught to Know What They Don't Know2 savers
• [2406.02528] Scalable MatMul-free Language Modeling1 savers
• The Markets Aren't Alright3 savers
• Why China got rich and India didn't - David Oks18 savers
• Automated Weak-to-Strong Researcher17 savers
• [2506.04645] Inference economics of language models1 savers
• The Dynamics of Commodity Spot and Futures Markets2 savers
• Continuous batching from first principles1 savers
• The Ultra-Scale Playbook - a Hugging Face Space by nanotron8 savers
• flowersforalgernon.pdf4 savers
• Introducing Atlas RF Studio: Toward a Foundation Model for Electromagnetics | ${title}4 savers
• nvidia-rtx-blackwell-gpu-architecture.pdf1 savers
• Worse Is Better2 savers
• New Age Commodities - Sean Cai1 savers
• [2212.07677] Transformers learn in-context by gradient descent6 savers
• Go smol or go home | Harm de Vries3 savers
• The TAM Is All of Human Labor - Google Docs1 savers
• AI Tools in Society: Impacts on Cognitive Offloading and the Future of Critical Thinking1 savers
• quantum-computation-and-quantum-information-nielsen-chuang.pdf1 savers
• Specialization and the Hive Mind1 savers
• Mini Project: GPU Accelerated Matrix Multiplication (almost) like cuBLAS1 savers
• 为什么线性注意力要加Short Conv？ - 科学空间|Scientific Spaces1 savers
• design manifestos .org15 savers
• Things unlearned4 savers
• What if RL Environments Aren't Mispriced?4 savers
• What happens when you run a CUDA kernel2 savers

highlights — 1

• The Rawlsian extreme is also a bad model, as the world doesn't wholly revolve around its single unhappiest inhabitant

  The Markets Aren't Alright