Rahul
@selfawareatom
Followers
2K
Following
6K
Media
140
Statuses
2K
Founding member and leading the foundation models team @sarvamai.
Joined August 2009
This is absolutely true. Never vibe code what you don't understand. If you do, then how will you know if the code's vibe matches with yours. This applies more to people just learning to code. Learning should be hard. You should spend dozens of hours debugging. That's how you.
"I use AI in a separate window. I don't enjoy Cursor or Windsurf, I can literally feel competence draining out of my fingers.". @dhh, the legendary programmer and creator of Ruby on Rails has the most beautiful and philosophical idea about what AI takes away from programmers.
3
6
116
RT @rasbt: Btw if you're learning how to build LLMs from the ground up, there's now a 17h companion video course for my LLMs From Scratch b….
0
285
0
You can also contribute to this and get it merged: However, the contribution should be conceptual rather than cosmetic.
0
0
13
Hi coders,. Implement Muon (easy), and make it work with TP, CP, PP (hard). If your PR gets merged into Megatron-LM, I will hire you on the spot. You have 7 days.
6
8
224
Long ago I found a backdoor in a famous news portal and slurped 40M+ news articles from their database. P.S.: Only used it for my research and promptly told them about it.
1
0
22
People might not realize this, but the translation models are the most important of all the models we build.
Congratulations to @SarvamAI on advancing AI-powered translation 🎊. Built with Gemma 3 4B-IT, Sarvam-Translate supports 22 Indian languages handling document translation, markdown, OCR, and more.
4
3
116
RT @indiainpixels: I've made a project called "Shruti" to preserve the memories from fading cultures of India and I invite you to contribut….
0
467
0
RT @ErnestRyu: New lecture recordings on RL+LLM! 📺. This spring, I gave a lecture series titled **Reinforcement Learning of Large Language….
0
151
0
RT @SarvamAI: The Sarvam developer experience just got an upgrade. We’ve made a few updates to the dashboard - . ◾️ A more intuitive playgr….
0
19
0
RT @ezyang: Want to play around with code that uses PyTorch DTensors but annoyed at having to go multiprocess? Make a fake process group! T….
0
28
0
I take more time than the whole video to just understand the question, and lil' bro spent one minute just trying to submit his solution 😭.
1
0
19
Here is the catch: it's 10,000 hours of "focused" work. Even people with a high attention span can only do 3-4 hours of such high-quality work per day.
BadMephisto on how he became great at speedcubing, AI research and teaching.
4
0
30
Prof @CKRaju14 has a nice talk about this:
holy shit, does this mean newton was not the first one to invent calculus?
0
0
8
Schwa deletion. It is called Schwa deletion.
It’s Ramayan, not Ramayana. It’s Ram, not Rama. We need to avoid the anglicization of our words. Our heritage doesn’t need a colonial accent. It’s written by Valmiki, FYI.
0
0
5
Ever since I watched the original LOTR trilogy, I have always wished Peter Jackson would bring the Ramayana and Mahabharata to the big screen. I want to see the 2 epics brought to life at the scale they deserve. Today's Ramayana teaser has rekindled my hope. Also, Hans Zimmer!.
1
2
32
