LogoThread Easy
  • 探索
  • 線程創作
LogoThread Easy

Twitter 線程的一站式夥伴

© 2025 Thread Easy All Rights Reserved.

探索

Newest first — browse tweet threads

Keep on to blur preview images; turn off to show them clearly

people keep asking me NONSTOP what a G1 shoulder socket looks like, so here you go

people keep asking me NONSTOP what a G1 shoulder socket looks like, so here you go

gpus and tractors Neural networks from Scratch book: https://t.co/hyMkWyUP7R https://t.co/8WGZRkUGsn

avatar for Harrison Kinsley
Harrison Kinsley
Wed Dec 10 22:34:28
at >$10 per mToks of inference, it seems plausible that SpaceX could operate orbital compute profitably, assuming it can stack Starship launch economics with Starlink for networking. The harder question imo isn’t “can this make money?” but “is it cheaper or more scalable than putting the same amount of compute in Texas?”

big unknowns for me are GPU avail/util and sat architecture. Depends heavily on rad-induced failures, derating, and how much redundancy you'd need(?); architecture comes down to the mass fractions for panels, radiators, structure, shielding, and silicon — ie, what the “platonic ideal” orbital compute bus looks like. 100kw per sat is awesome if it could get than lean, could get to 15-30 MW compute per launch.

if you assume exponential compute demand and linear terrestrial datacenter buildout (power, permitting, construction, etc.), the case for orbital compute gets stronger over time. few are better positioned to take this view than elon. in that world, one could imagine mass-manufacturing compute clusters you can launch and power with ~95% capacity factor space solar starts to look like the fastest way to scale. 

TL;DR: let elon cook

at >$10 per mToks of inference, it seems plausible that SpaceX could operate orbital compute profitably, assuming it can stack Starship launch economics with Starlink for networking. The harder question imo isn’t “can this make money?” but “is it cheaper or more scalable than putting the same amount of compute in Texas?” big unknowns for me are GPU avail/util and sat architecture. Depends heavily on rad-induced failures, derating, and how much redundancy you'd need(?); architecture comes down to the mass fractions for panels, radiators, structure, shielding, and silicon — ie, what the “platonic ideal” orbital compute bus looks like. 100kw per sat is awesome if it could get than lean, could get to 15-30 MW compute per launch. if you assume exponential compute demand and linear terrestrial datacenter buildout (power, permitting, construction, etc.), the case for orbital compute gets stronger over time. few are better positioned to take this view than elon. in that world, one could imagine mass-manufacturing compute clusters you can launch and power with ~95% capacity factor space solar starts to look like the fastest way to scale. TL;DR: let elon cook

Hmm

avatar for Ryan McEntush
Ryan McEntush
Wed Dec 10 22:34:13
RT @SVenkatachalam: Indeed, the future of interfacing with the web is humans not having to do it at all!! Congrats @yutori_ai

RT @SVenkatachalam: Indeed, the future of interfacing with the web is humans not having to do it at all!! Congrats @yutori_ai

Co-founder & Chief Scientist @yutori_ai. Prev: Senior Director leading FAIR Embodied AI @MetaAI and Professor @GeorgiaTech.

avatar for Dhruv Batra
Dhruv Batra
Wed Dec 10 22:33:50
RT @s_mohinii: 7/Affordable stainless steel cookware

Find high-quality stainless-steel cookware under strict criteria
Minimum 3 pieces, un…

RT @s_mohinii: 7/Affordable stainless steel cookware Find high-quality stainless-steel cookware under strict criteria Minimum 3 pieces, un…

Co-founder & Chief Scientist @yutori_ai. Prev: Senior Director leading FAIR Embodied AI @MetaAI and Professor @GeorgiaTech.

avatar for Dhruv Batra
Dhruv Batra
Wed Dec 10 22:32:19
RT @s_mohinii: 6/ Monitor vintage Chanel with precise filters

Vintage Chanel women’s pieces under $1,000 — alerting you the moment new lis…

RT @s_mohinii: 6/ Monitor vintage Chanel with precise filters Vintage Chanel women’s pieces under $1,000 — alerting you the moment new lis…

Co-founder & Chief Scientist @yutori_ai. Prev: Senior Director leading FAIR Embodied AI @MetaAI and Professor @GeorgiaTech.

avatar for Dhruv Batra
Dhruv Batra
Wed Dec 10 22:32:15
RT @s_mohinii: 5/As I kept using it, I realized travel is only one slice of what Scouts can handle.

Here are three  other  things Scouts d…

RT @s_mohinii: 5/As I kept using it, I realized travel is only one slice of what Scouts can handle. Here are three other things Scouts d…

Co-founder & Chief Scientist @yutori_ai. Prev: Senior Director leading FAIR Embodied AI @MetaAI and Professor @GeorgiaTech.

avatar for Dhruv Batra
Dhruv Batra
Wed Dec 10 22:32:11
  • Previous
  • 1
  • More pages
  • 1053
  • 1054
  • 1055
  • More pages
  • 5634
  • Next