Hemanth @hemtejay

Joined February 2022

Tweets

10
Followers

15
Following

536
Likes

412

Hemanth @hemtejay

a month ago

With Apple’s new A20 chip breakthrough with 2nm chips, are we heading into a world where the on-device local routers treat local NPU cores and distant cloud GPUs as different expert nodes within the exact same inference pass ?

0 0 1 33 0

View Details

Hemanth @hemtejay

a month ago

Apple foldableee !!!

Marques Brownlee @MKBHD

a month ago

One More Thing

130 185 8K 534K 64

0 0 2 165 0

View Details

Hemanth @hemtejay

a month ago

@johnternus @tim_cook Variable aperture ✨

0 0 0 49 0

View Details

Hemanth @hemtejay

2 years ago

@dhruv_rathee @AravSrinivas Aren’t they primarily a search first company?

0 0 10 3K 0

View Details

Aravind Srinivas @AravSrinivas

2 years ago

DeepSeek-R1-Zero (no supervised fine-tuning) showing human-like reasoning skills in natural language just by virtue of reinforcement learning (RL).

87 283 3K 265K 685

View Details

Aravind Srinivas @AravSrinivas

2 years ago

There’s a lot of misconception that China “just cloned” the outputs of openai. This is far from true and reflects incomplete understanding of how these models are trained in the first place. DeepSeek R1 has figured out RL finetuning. They wrote a whole paper on this topic called DeepSeek R1 Zero, where no SFT was used. And then combined it with some SFT to add domain knowledge with good rejection sampling (aka filtering). The main reason it’s so good is it learned reasoning from scratch rather than imitating other humans or models.

232 1K 9K 983K 2K

View Details

ℏεsam @Hesamation

2 years ago

Respect! Hugging Face 🤗 is reproducing the whole DeepSeek R1 pipeline to be used by open source community. So far it has > GRPO implementation > train and evaluation code > generator for synthetic data