John Rose

@CathPoaster Relatable. Have you considered naming your stack

right monitor is 20 codex instances. left monitor has situational awareness on autoscroll. center monitor is my word doc mainfesto. two keyboards, one for both hands. left airpod is dwarkesh x eric jang, 3x speed. right airpod tchaikovsky. meta quest 3 overlays my HUD: heart rate, words per minute, blood caffeine content. one assistant hooks me to an iv of chinese peptides, cocktail. the other feeds me kimchi. my unitree robot steps in when my posture slouches. blue light beams down on me in my herman miller chair. efficiency. no wasted movement. no wasted thoughts. think you can keep up with me? good luck. this is just for my morning emails.

@ruima Disclaimer: wild speculation

@ruima If all TPUs, v5e assume 5000 tok/s, total ~12k TPUs at 100% util, or ~24k TPUs reqd with traffic/redun

Wait what? These guys are saying Claude Code only has 14mm DAU?? Interview from 2 days ago Living in SV and working in tech (basically?) really distorts my sense of normal sometimes

@jm_alexia Longer training offset by parallel/modular potential (maybe)

Built Sakana Diffusion Blocks into a language model

@conorsen Same. Compute limitation or flexing pricing power?

I burned through all my tokens in a session on Claude Pro this morning in maybe 10 minutes trying to pull data out of one PDF — there’s just no way there’s enough compute to disrupt a meaningful number of jobs this year.

@m_takewaka Texas. No debate

My American friends, I'm facing a major problem here. When I say I want to eat American BBQ, some Americans say, "come to Texas," others say, "come to South Carolina," and still others say, "come to Missouri." I'm confused. Are you guys going to start a BBQ civil war?

@DanielBogdanoff Systems eng, wrote Linux PCI drivers that still ship 20 yrs later. Amassing ESP32s for world domination

Electrical Engineering Twitter Role Call! If you want a feed full of EE/electronics go though and connect with all the profiles that respond. Add yourself -> post what you do in < 5 words (stealing this idea from⬇️, I love it!)

@hill_cape Incredible. Howdy from Texas

個展まで残り６日となりました。 最後の追い込みです。。 デザフェスなどではバタバタしててゆっくり空中配線を楽しめなかった方も、個展ではじっくりゆっくり楽しめます。 撮影・SNS大歓迎です。 おそらくここ以外では見られない、至高の空中配線を大量摂取できる、空中配線展にぜひお越しください。 「柊工房　空中配線展」 東京都渋谷区神宮前3-20-18 デザインフェスタギャラリー原宿 WEST-2A 全日在廊 入場無料 3/20(金) 11:00-19:00 3/21(土) 11:00-19:00 3/22(日) 11:00-16:00

built Paged Attention from scratch in CUDA. how does vLLM manage precious KV Cache in GPU memory? it borrows from OS-level memory paging concepts see my artisanal code below

HBM getting expensive, better inspect it.