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! ๐งต
Blog:
I open-sourced my Claude Code HUD.
It shows how full the context is, the state of my repos, and how long the current task will take. When I come back from a break, it tells me where the session stopped.
Less context switching. Lower cortisol.
One thing I realized after watching this video: most founders (me included) swing between two extremes. Either constantly filtering and finding new creators, or only trusting a very few.
Both are traps. With the former, you never build trust. With the latter, creators get too much bargaining power and get slack.
The sweet spot is in between: keep searching for new ones, but build strong trust and a culture bond with the best matches.
literal hardest role to hire for rn and every ai startup wants someone
> chronically online (knows trends)
> has taste & can create (new media)
> understands ai & technical
> can execute with quality (operator)
such a small niche of folks who can do this
literal hardest role to hire for rn and every ai startup wants someone
> chronically online (knows trends)
> has taste & can create (new media)
> understands ai & technical
> can execute with quality (operator)
such a small niche of folks who can do this
I am shocked... My daughter took my Luna Ultra these couple of days and I have no clue what photos she took until today. THEY ARE CRAZY GOOD (zero edits!) ๐คฏ
She is a brilliant photographer