Chrizhuu 🦔🍩
Verified account Sonic SVM@Chrizhuu
5.3K Following 9.5K Followers 4.9K Posts
Latest posts 5
Been lookin at gacha e-commerce products like icybox, courtyard and others for a bit
Clicked today that they are offering a way to spin a wheel to gain an item of respect.
This is especially true for lower income bracket, because when you own very little, you care about being respected by others the most. The watches, shoes or bags are symbol of wealth not for wealth’s sake but for the respect it holds.
Perps functions similarly but it’s greater risk and doesn’t hold “respect” the same way a nice pair of sneakers or a nice expensive watch does
Clicked today that they are offering a way to spin a wheel to gain an item of respect.
This is especially true for lower income bracket, because when you own very little, you care about being respected by others the most. The watches, shoes or bags are symbol of wealth not for wealth’s sake but for the respect it holds.
Perps functions similarly but it’s greater risk and doesn’t hold “respect” the same way a nice pair of sneakers or a nice expensive watch does
Rewatching Claude’s “Keep Thinking” video from last year.
Crazy to think Claude Code only launched last year.
Every morning I'm reminded how valuable LLMs are for lowering the friction of “starting to think.”
Sometimes I wonder whether removing that initial friction weakens the muscle that gets me thinking in the first place. But the productivity gains have been immense.
Think this video puts it beautifully.
youtube.com/watch?v=FDNkDB…
Crazy to think Claude Code only launched last year.
Every morning I'm reminded how valuable LLMs are for lowering the friction of “starting to think.”
Sometimes I wonder whether removing that initial friction weakens the muscle that gets me thinking in the first place. But the productivity gains have been immense.
Think this video puts it beautifully.
youtube.com/watch?v=FDNkDB…
Black magic in the modern age
if you have a qwen3.5 4b around, please try steering neurons L14 N8949 +1 and L13 N7956 -1View quoted post on X
now you have evil qwen
recipe: dose × median(‖resid‖ leaving layer L, tokens 1:) / ‖W_down[:, n]‖` at the down_proj at every input