09:03:34

Why does matrix vector product work so well for finding similarity between the text features and the image features in clip embeddings? compared to K nearest neighbors, which works great for finding image features that are similar to other image features but doesn't work for text

🗨️ 1 ♺ 0 🤍 10


09:11:34

the answer is: I WAS WRONG! it works just fine. i had a variable lying around in the jupyter kernel and didn't notice that i was calling the wrong thing hahaha

🗨️ 0 ♺ 0 🤍 6


10:39:06

okay i got Annoy nearest-neighbor search working for my meme search project. results are promising. tested on multiple image folders locally, vs. GPU-enabled matrix vector multiplication.

🗨️ 1 ♺ 0 🤍 8


10:39:07

it's also much faster to load once cached. sucks to take 2.5 minutes to process and index 26,000 images, but you only have to do it once (unless you add or remove from the dataset lol)

🗨️ 2 ♺ 0 🤍 4


10:39:07

over a 50 image dataset, Annoy is 3x slower building the index, and equal or slower on inference. over a 6,000 image dataset, it's 2x slower building index but over 30x faster at inf. over a 26,000 image dataset: 5x slower to build, but over 100x faster to infer! 50ms to 7secs!

🗨️ 1 ♺ 0 🤍 4


11:04:55

good mourning

🗨️ 1 ♺ 0 🤍 3


11:17:35

The woman from Brazil screaming "What did you do with his body?!" but it's just me every single day screaming "what did I do with my wallet"

🗨️ 2 ♺ 0 🤍 7


11:20:53

The answer, as usual: it was in my backpack right where it goes

🗨️ 0 ♺ 0 🤍 3


13:06:39

0 is a subprime number

🗨️ 0 ♺ 0 🤍 7


13:24:57

Not sure what I meant to type but it was not "restore" the post rats. it was probably "escape". if only y'all would stop being so smart and interesting and good at talking about things 🤬

🗨️ 0 ♺ 0 🤍 5


13:27:32

I left some Taco Bell sitting in the front seat of my car overnight. it's been 70s high and 40s low lately. is it still safe to eat? please let me know soon I'm very hungry

🗨️ 15 ♺ 0 🤍 11


13:33:46

The problem with being the one to do things before they are cool, is that if you're consistent about it, you're never cool

🗨️ 10 ♺ 4 🤍 71


13:51:20

i never turn down a fight to the death https://t.co/EA5ss6qEbn

🗨️ 0 ♺ 0 🤍 7


13:52:15

if you ever have to apologize to somebody but you don't want to, this trick always works. as long as they can't see your hand

🗨️ 0 ♺ 0 🤍 6


14:00:30

Lot of mfers out here Building a second brain, when they should really be Getting some first brain

🗨️ 6 ♺ 0 🤍 48


15:01:18

I'm the blond Shingy

🗨️ 0 ♺ 0 🤍 0


15:43:15

This thought is brought to you by all my friends are now learning Python gang

🗨️ 0 ♺ 0 🤍 11


15:46:05

broke: not having money woke: not having oppression bespoke: just the best poke

🗨️ 1 ♺ 0 🤍 6


15:50:28

People are always asking me this question. My answer is no

🗨️ 1 ♺ 0 🤍 25


17:00:36

i guess this does imply that I'm going to eat it... sooner or later https://t.co/C4xEkeYrJo

🗨️ 0 ♺ 0 🤍 1


17:04:30

literally get this Q on the reg, because of my display name. here's a recent answer from DMs https://t.co/rvYWNb66qu

image from twitter

🗨️ 2 ♺ 2 🤍 12


17:34:40

speaking of which, i am now accepting job offers for high-paying positions in beautiful cities. i have innumerable skills and I'm a joy to be around. fully remote teams are a possiblity but not preferred

🗨️ 0 ♺ 0 🤍 8


18:29:15

I spent roughly 15 grand on college stuff in 2010/2011. How much money would I have now if I had invested that into Bitcoin instead? https://t.co/JXHeRfDZq9

🗨️ 2 ♺ 0 🤍 9


18:33:57

The answer is roughly a quarter of a billion dollars. for anyone who is playing along at home

🗨️ 1 ♺ 0 🤍 8


19:12:00

Kanye Reeves

🗨️ 4 ♺ 0 🤍 14


20:10:13

The answer is actually I would have spent it when it reached a hundred bucks and felt awesome

🗨️ 0 ♺ 0 🤍 6


20:34:52

boo

🗨️ 3 ♺ 0 🤍 11


20:37:39

conservatism sucks. besides the fact that it doesn't make any sense in a world at the brink of catastrophic change, it's just BORING

🗨️ 2 ♺ 0 🤍 9


21:13:16

punching myself in the head as an extreme form of stimming

🗨️ 4 ♺ 0 🤍 27