Naveen Rao (@naveengrao) 's Twitter Profile
Naveen Rao

@naveengrao

VP GenAI @Databricks. Former CEO/cofounder MosaicML & Nervana/IntelAI. Neuro + CS. I like to build stuff that will eventually learn how to build other stuff.

ID: 22113265

linkhttps://github.com/mosaicml/composer calendar_today27-02-2009 05:53:32

4,4K Tweet

29,29K Takipçi

809 Takip Edilen

Naveen Rao (@naveengrao) 's Twitter Profile Photo

Cutting through the noise: which models work best for RAG and how does long context help? I think this is the most comprehensive analysis of model RAG performance. We also look at model failures! This is a great guide for users on what works for their application!

Naveen Rao (@naveengrao) 's Twitter Profile Photo

This is important for anyone trying to understand where we are in AI today. There are lots of amazing problems to solve…and it’s not *only * about scale. We need more fundamental breakthroughs. I’m here for that :)

Databricks Mosaic Research (@dbrxmosaicai) 's Twitter Profile Photo

Function calling significantly enhances the utility of LLMs in real-world applications; however, evaluating and improving this capability isn't easy — and no one benchmark tells the whole story. Learn more about our approach in the latest blog from Databricks:

Function calling significantly enhances the utility of LLMs in real-world applications; however, evaluating and improving this capability isn't easy — and no one benchmark tells the whole story. Learn more about our approach in the latest blog from <a href="/databricks/">Databricks</a>:
Naveen Rao (@naveengrao) 's Twitter Profile Photo

Demystifying function calling! As part of compound systems, calling funcs/APIs is super important. There's still a fair bit of work required to make these reliable. Learn about it here 👇 databricks.com/blog/unpacking…

Naveen Rao (@naveengrao) 's Twitter Profile Photo

Some great work by one of our undergrad interns! We are constantly looking for ways to improve quality of outputs leveraging existing artifacts like open LLMs. RLHF has been shown to work but is expensive. Here we show a technique to leverage the implicit capabilities of models

Naveen Rao (@naveengrao) 's Twitter Profile Photo

We've been working with Twelve Labs (twelvelabs.io) for more than a year now. These guys took a different strategy than most other AI startups and focused on gen AI and video. That bet is paying off now! Think "RAG but for video"

Naveen Rao (@naveengrao) 's Twitter Profile Photo

I posted this nearly 2 years about Tesla’s Dojo chip. Many Tesla fans argued that i was underestimating what they could do. Well, Elon Musk just disclosed his 100k H100 cluster built in 4 months.(!) As I said then, it’ll take years for a new hardware platform to actually work