vani

some day i’ll have a bay club sf membership and it will be so well-deserved

someday they’ll make a highway/driving white noise sound and it will lull ppl to sleep

@jijijooeun @daymakerguy lies

SF is so stuck up on AI. I literally sell cakes and I make more than every single one of you

@yacineMTB rent a gpu on llama, then ssh or colab lololol

can i run cuda on a macbook

the gen z stare to me is “am i stupid or are you stupid” like when some guy asked me what the speed of gravity was

spent 45 mins searching for my first ml project demo video and finally found it and it's literally just "Demo.mp4" like yeah trust 16 yo me to call it that tbh i would probably still call it that

all i want is to learn some more physics

thinking back to when i was freshman and my senior friends dared me to do the stanky leg for their entertainment

this is so cool

“do not reply to this email” but i really wanted to send you a super enthusiastic thank you

there is no greater feeling than freeing yourself from someone or something that wasn’t meant for you

Energy creates energy, start moving.

am i horribly unskilled or is c syntax harder than cuda

born to have a huge wall-mounted whiteboard, forced to scribble everything into my graph-ruled notebook

@ProofofMaro mmm crunchy

She’s a 10 but she thinks raspberry pi is dessert

@fdotinc super secret code? (:p)

@vaniguptaaaa luma.com/worldsinaction…

world models can u use them to monitor the situation? idk but we're throwing the first world model hackathon where u will get access to: -workshops to catch u up to speed -world labs, genie & pico vr -mentors & judges from ByteDance, Meta & Google comment MODEL for an invite

crashed out and i am finally so back

@jino_rohit im doing something similar! excited to see how it goes for you

early last year i did a bit of gpu programming, i wrote a kernel for 30 days straight and then went off exploring agents and tons of open source. this helped me navigate large codebases and overall made me a better programmer. now i'm back at ml systems again. talked to a bunch of smart people, took some advice, and this is roughly what i want to, starting a 90 day challenge and splitting most days roughly in two - 1. gpu programming - raw cuda kernels mixed with cute dsl. already spent some time on cute so i have a small headstart. the goal is to mix both so i really feel why cute makes sense and how to write better kernels by iterating between the two. 2. inference stuff - reading ml systems papers, tricks, then actually implementing them myself. ive also already spent some time here so i can push into more real codebases like vllm, sglang, etc. some days might be harder to follow, but the point is to stay consistent and keep myself accountable by putting it out there.

bad bunny duolingo avatar who teaches you spanish