i turn coffee into gpu bill
Joined December 2022
- Tweets2.6K
- Following1K
- Followers1.6K
- Likes30.7K
you need to post the rent in dollars if you're going to ask for 90k for a studio apartment in bangalore
India/US cross border at its peak - saw a bangalore housing post in sf housing group chat 🤣💀@peakbengaluru
whenever i see someone being over optimistic about india, it's either someone who doesn't live here or is a vc who needs to justify their existence
jackson lamb is the closest thing to dr. house on tv right now, i had given up on seeing a character like him again
you're absolutely right, hiring an assassin from deep web to shoot root in the knees was unethical, i should not have done that
tbh i respect the people who built the models because even building whatever they did with a very small young team with no one who has been there done that before is impressive
they hardly had a dozen people working on building models, looks like many of them are leaving?
After 2.5 years at Sarvam, I am starting something new!
I had a great time leading the RL team. We launched the 30B and 105B models back in Feb with a young team building everything from scratch. Excited to see the team keep growing and pushing the frontier. Grateful to @pratykumar for the early bet and trust.
As for what's next: I am convinced that there are a few critical parts of the stack that we still need to build to meaningfully accelerate science. One of them is verification.
Scientific discovery only scales with compute when proposing and verifying new ideas is fast and cheap. In math and code, we're starting to see what happens when models can generate ideas and reliably check them at scale. OpenAI's recent Navier-Stokes work is a glimpse of what that looks like.
Similar moments in chemistry, materials, biology and the rest of the physical sciences would require verification to get dramatically faster and cheaper. It cannot be bottlenecked by the throughput of physical labs
More soon. Very excited for what's next.
it works tbh, how dare you try it on any dataset beyond the two pointless ones every paper shows it works with
I was so excited until I read the last sentence.
My honest read is that the model doesn't actually communicate more naturally; it gets caught with the smoking gun less often.
