harshith @theharshithh

inference @simplismartHQ, prev @onfinance_ai github.com/theharshithh Joined April 2022

Tweets

594
Followers

314
Following

3K
Likes

3K

harshith @theharshithh

3 days ago

its insane. u guys really understood taste. demo videos like this is what we want!

Capy @capydotai

3 days ago

its insane. u guys really understood taste. demo videos like this is what we want!

53 51 588 71K 309

Download Video

0 0 2 69 0

Prime Intellect @PrimeIntellect

3 weeks ago

10 9 182 20K 9

Download Image

harshith @theharshithh

6 days ago

speculative decoding is hard. fking hard. easy to underestimate at the first sight

1 0 3 80 0

harshith @theharshithh

6 days ago

there is no going back after using whispr. such a clean product.

0 0 0 31 0

1. codex is insane, much better than claude code with gpt5. 2. how do u sleep at night knowing i can never continue the same conversation on codex as there is no way i could give current session id later on to continue that context it's like I spend hrs getting context warm,…

Sam Altman @sama

2 weeks ago

448 52 2K 375K 140

0 0 1 221 0

vik @vikhyatk

a week ago

9 32 791 47K 57

Download Image

elie @eliebakouch

a week ago

The technical report of @Meituan_LongCat LongCat-Flash is crazy good and full of novelty. The model is a 560B passive ~27B active MoE with adaptive number of active parameters depending on the context thanks to the Zero-Computational expert. 1) New architecture > Layers have 2…

23 153 867 240K 757

Download Image

harshith @theharshithh

a week ago

ray is one beautifully written software. gone are the days where packages are so elegant.

1 0 1 33 0

Yuchen Jin @Yuchenj_UW

3 weeks ago

coining this as Jin's law

6 3 74 7K 1

harshith @theharshithh

3 weeks ago

for those who don't know, i joined @SimplismartHQ a few months ago to have the best training speeds(faster than unsloth) and inference throughput.

elie @eliebakouch

3 weeks ago

for those who don't know, i joined @SimplismartHQ a few months ago to have the best training speeds(faster than unsloth) and inference throughput.

7 4 145 25K 11

0 0 5 66 0

Omar Khattab @lateinteraction

3 weeks ago

Every once in a while, it hits you that word2vec and attention were one year apart.

19 36 1K 109K 136

harshith @theharshithh

3 weeks ago

also checkout simplismart.ai, they just got a beautiful makeover!

0 0 2 37 0

Jack Morris @jxmnop

4 weeks ago

OpenAI hasn’t open-sourced a base model since GPT-2 in 2019. they recently released GPT-OSS, which is reasoning-only... or is it? turns out that underneath the surface, there is still a strong base model. so we extracted it. introducing gpt-oss-20b-base 🧵

159 470 6K 919K 4K

Download Image

harshith @theharshithh

4 weeks ago

oai and gpt5 have successfully cracked the code for a useful and yet fast coding model. this is the sonnet 3.5 moment

0 0 1 63 0

harshith @theharshithh

4 weeks ago

"a kid born today will never be smart than AI" - @sama this single thing hit me so hard about the progress we had made from the last 2 years

0 0 1 84 1

tokenbender @tokenbender

a month ago

i am free of my curse (momentarily). pure RL in pretrain from random weights works. now successfully scaling generally on natural language. writing everything, would share later.

19 15 264 16K 68

Download Image

harshith @theharshithh

a month ago

finally now i need @garrytan's rescue. letsss goo!

1 1 6 132 0

Download Image

harshith @theharshithh

a month ago

guys, a friend of mine is looking for a frontend dev for a fast-paced startup. they are backed by solid investors who are knees deep in the domain. dm if anyones interested.