vik @vikhyatk

teaching computers how to see // prev: @awscloud moondream.ai Seattle Joined November 2008

Tweets

3K
Followers

7K
Following

518
Likes

13K

miru (mul/acc) @miru_why

19 hours ago

know the difference

9 21 307 8K 21

Download Image

vik @vikhyatk

20 hours ago

tired of python. i'm going to write a machine learning framework in ruby. i need to get my groove back

8 0 47 4K 1

vik @vikhyatk

21 hours ago

told my friend I was gonna spend the day replicating some arxiv paper and she said "what a rich and fulfilling life you lead"

11 2 144 10K 9

vik @vikhyatk

22 hours ago

i have a lot of respect for how anthropic openly shares their interpretability research. now if you’ll excuse me i’m off to try and train some sparse autoencoders

Adam Jermyn @AdamSJermyn

2 days ago

i have a lot of respect for how anthropic openly shares their interpretability research. now if you’ll excuse me i’m off to try and train some sparse autoencoders

1 16 115 71K 90

0 0 30 2K 6

vik @vikhyatk

23 hours ago

the audacity of calling elon a climate change denier when he’s singlehandedly accelerated the world’s timeline for decarbonization more than any government out there

4 2 35 2K 1

i was a good citizen for years, i make ONE mistake and they take away my coding license because i used exceptions for control flow. why can't the seattle police department focus on real crimes instead?

4 1 36 3K 1

vik @vikhyatk

a day ago

fear not the person who has trained 10,000 models once, fear the person who has trained one model 10,000 times

4 8 132 7K 5

vik @vikhyatk

2 days ago

so fast you don't even need to implement streaming text generation

cj @cj_pais

2 days ago

so fast you don't even need to implement streaming text generation

3 1 34 17K 16

Download Video

4 2 51 7K 9

vik @vikhyatk

2 days ago

corollary: there is still a lot of alpha in building models optimized for real word use cases rather than MMMU

typedfemale @typedfemale

2 days ago

corollary: there is still a lot of alpha in building models optimized for real word use cases rather than MMMU

7 3 108 16K 18

0 0 29 2K 1

vik @vikhyatk

2 days ago

if they let me onto this board i will force openai to tell us when GPT-5 is launching

vik @vikhyatk

2 days ago

if they let me onto this board i will force openai to tell us when GPT-5 is launching

0 0 23 4K 0

1 0 22 2K 0

vik @vikhyatk

2 days ago

you wouldn't last an hour in the asylum where they raised me

3 2 32 2K 1

Download Image

vik @vikhyatk

2 days ago

this is probably the easiest way run moondream right now. just download it and run ./moondream2.llamafile to start the llama.cpp server chmod +x first if you're on mac/linux

cj @cj_pais

2 days ago

this is probably the easiest way run moondream right now. just download it and run ./moondream2.llamafile to start the llama.cpp server chmod +x first if you're on mac/linux

2 7 54 5K 31

0 0 22 2K 12

vik @vikhyatk

2 days ago

dario amodei wants me to delete this tweet because it discloses a compute multiplier, but i will not be silenced 😡 instead i will tell you that scaling by a factor of 4 instead of 8 will likely work even better