Y Hacker News new | ask | show | jobs

user: jxmorris12 created: 2016-12-07 karma: 4843 personal website: jxmo.io submissions: Gram Newton-Schulz: A Fast, Hardware-Aware Newton-Schulz Algorithm for Muon 17 points | 3 comments 0 points | 0 comments The Decline of Token-Level Purchasing Power 2 points | 0 comments 0 points | 0 comments A one-parameter model that gets 100% on ARC-AGI-2 3 points | 0 comments The Effective Sample Size 25 points | 1 comments Can you go 82-0? 5 points | 0 comments Automating Plain-Text Location Updates with Apple Shortcuts and Redis 2 points | 0 comments Bias Compounds, Variance Washes Out 29 points | 32 comments 0 points | 0 comments Rothko for your current weather conditions 151 points | 15 comments When AI Starts Writing Systems Code 1 points | 0 comments DTensor, Correctness and the Costs of Abstraction 2 points | 0 comments The just-say-no engineer was a ZIRP phenomenon 160 points | 131 comments The AI Decoupling 2 points | 0 comments The Lottery – Shirley Jackson (1948) 41 points | 55 comments Yagni 2 points | 0 comments LoRA and Weight Decay (2023) 50 points | 2 comments Through the looking glass of benchmark hacking 34 points | 15 comments Diversity as the Bottleneck in Self-Play 2 points | 0 comments The Moth Story Map 35 points | 3 comments Boosting multimodal inference performance by >10% with a single Python dict 16 points | 0 comments Why are neural networks and cryptographic ciphers so similar? (2025) 143 points | 48 comments Using group theory to explore the space of positional encodings for attention 48 points | 0 comments Pyptx – A Python DSL to Write Nvidia PTX for Hopper and Blackwell 3 points | 0 comments Which one is more important: more parameters or more computation? (2021) 59 points | 13 comments As We May Think 1 points | 0 comments Work with the garage door up (2024) 202 points | 127 comments Six (and a half) intuitions for KL divergence 114 points | 33 comments Thoughts on AI and Research [pdf] 3 points | 0 comments