We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.
We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵
Blog: https://t.co/bOWCcFhCVq
@theo Oh man, I was working on something related to that for one of my companies, and I reduced the time from 24 minutes to about 3 minutes using a bunch of caching strategies, along with Rust and Go.
What's New in Grok Bot
00:00 🎃
00:13 Pages in Grok Bot
01:09 Vault
02:05 1Password
02:21 Apple integrations
03:21 Bot makes videos
03:47 New bot characters
04:13 X Spaces
04:31 Font fun
04:54 Design polish
05:11 Rapid fire