正在加载视频...

视频加载失败

Codex computer use + Spark absolutely ripping through Universal Paperclips:

51,945 次观看 • 5 个月前 •via X (Twitter)

20 条评论

Lee Fowler 的头像
Lee Fowler5 个月前

> Me watching Codex Computer Use burn through all my Fusing Orbs in PoE at mach 17 and still no 6-link

Sughu 的头像
Sughu5 个月前

is this sped up?

leo 的头像
leo5 个月前

it's realtime!

Sughu 的头像
Sughu5 个月前

wtf

Alexander Temerev 的头像
Alexander Temerev5 个月前

Had to beg for more wire. Pathetic.

Matt 的头像
Matt5 个月前

universal paperclips is actually a decent stress test, pure click loops with escalating complexity. curious how it handles something that needs switching between 3 apps while keeping state, that's where every computer use demo I've tried falls apart

Muki 的头像
Muki5 个月前

Probably a very good target for autoresearch since you can measure milestones for the speedrun.. human record is under 2 hours..

banteg 的头像
banteg5 个月前

this is fun, tried it too. even with gpt-5.4, it's very stupid in the early game. mine just jacked up prices and sit on 90% of unsold inventory.

Mina Fahmi 的头像
Mina Fahmi5 个月前

this is the optimal eval

omni 的头像
omni5 个月前

Looks like a human with codex spark. What reasoning effort is this? Is it just as fast with GPT-5.4 fast mode?

Lev 的头像
Lev5 个月前

Cool!

Sonya.tsv 的头像
Sonya.tsv5 个月前

This was exactly how I noticed that we are not really given that much Spark quota, burned through weekly 10% in a couple of minutes

LinearUncle 的头像
LinearUncle5 个月前

can u test ComputerUse + figma, and ask codex to design a login page in figma?

alakazam 1314 的头像
alakazam 13145 个月前

whoa this is cool

Adel Bucetta 的头像
Adel Bucetta5 个月前

looks like someone's trying to automate away their own existential crisis

Stephen 的头像
Stephen5 个月前

more amazing that you got it to work, cannot get spark to use computer use

Xbotter 的头像
Xbotter5 个月前

Spark's speed is unbelievable.

SaltBAE Systems 的头像
SaltBAE Systems5 个月前

It's terrible at the game though?

V0id 的头像
V0id5 个月前

why not just use agent browser?

Kelvin聊AI落地 的头像
Kelvin聊AI落地5 个月前

I don't see spark model in codex , why?

相关视频