zayon @bootcoff

I am retard. only now did i realize that you can immediately "guess" ddim from the forward process by "splitting" the noise. Then you just replace X_0 and X_1 by their predictions. smh

0 0 1 41 1

Download Image

DEJAN @dejanseo

a month ago

@xikun_zhang_ The bubble is strong in this one.

0 1 0 534 0

zayon @bootcoff

2 months ago

I can’t believe that some people thought tutorial at icml = a talk about all your work Smh

0 0 0 83 0

Julian Lehr @julianlehr

2 months ago

First-mover advantage doesn't mean you can't catch up on product. It means you can't catch up on distribution despite delivering an equally good or better product.

Nikita Bier @nikitabier

2 months ago

First-mover advantage doesn't mean you can't catch up on product. It means you can't catch up on distribution despite delivering an equally good or better product.

344 218 5K 484K 448

Download Image

27 26 666 67K 116

LLM RL code does not need to be complicated! Here is a minimal implementation of GRPO/PPO on Qwen3, from-scratch in JAX in around 400 core lines of code. The repo is designed to be hackable and prioritize ease-of-understanding for research: github.com/kvfrans/lmpo

11 46 449 30K 472

Download Image

Joseph Alessio @alessio_joseph

2 months ago

@tszzl racism learning through hitlerian feedback

4 7 220 12K 7

zayon @bootcoff

2 months ago

i am discovering the literature on the analysis of Best-of-N and related methods and my first struggle is this: this is simply sampling importance resampling, why do we name if soft best-of-N now?

0 0 2 84 0

Download Image

Nate Silver @NateSilver538

2 months ago

Just a little casual racism from Silicon Valley's finest.

Shaun Maguire @shaunmmaguire

2 months ago

Just a little casual racism from Silicon Valley's finest.

2K 2K 15K 5.7M 942

Download Image

1K 566 9K 1.7M 384

zayon @bootcoff

2 months ago

Great work but man the apple and the comparison to Galileo lmaooooo

Zeyuan Allen-Zhu, Sc.D. @ZeyuanAllenZhu

4 months ago

Great work but man the apple and the comparison to Galileo lmaooooo

23 170 1K 334K 1K

Download Image

0 0 1 84 0

🦋/acc 🌲 @argyros_selini

2 months ago

Never give up on your dreams If you work hard enough & help lay down the foundations of AI by creating backpropagation, CNNs, become directory of NYU & win the Turing Award. One day you might be able to work under someone who started a digital sweatshop company 😀

55 134 4K 281K 562

Download Image

Quentin @qtnx_

2 months ago

"you don’t seem to understand, i have a PhD in ML, i was meant to pretrain language model" "wrap the fucking API"

32 183 4K 165K 326

Bradley Dorr @CarniContrarian

2 months ago

@alexxsedlak Online presence is generally for insecure people that can’t function without constant reassurance from strangers flexing online. It’s really sad to watch. Just go do things.

0 1 54 942 2

John Hudson @John_Hudson

2 months ago

SCOOP: The United States obtained intercepted communication between senior Iranian officials discussing this month’s military strikes on Iran’s nuclear program and remarking that the attack was less devastating than they had expected, per four soures