saurabh_here1's profile picture. Video AI researcher @TheCantinaApp. Led the development of one of the world’s fastest talking avatar model, try at http://cantina.ai. prev @iitdelhi @uw @UofIllinois.

Saurabh Shukla

@saurabh_here1

Video AI researcher @TheCantinaApp. Led the development of one of the world’s fastest talking avatar model, try at http://cantina.ai. prev @iitdelhi @uw @UofIllinois.

If you are learning anything new, try to go really deep. Fundamental level deep. Real fun starts when you are at a certain level of understanding and more depth gives you goosebumps. Anyone else felt it?


Its a classic textbook case of how Apple missed the AI train, despite having cash as well as talent.

BREAKING: Apple finalizing deal that pays Google $1 billion a year to integrate Gemini into Siri



1. Build a powerful model. 2. Integrate it with all your hit products. 3. Win the space. Google’ recipe of winning the AI race.

🧵 New feature drops 🧵 Navigation in Google Maps is getting a powerful boost with Gemini. Ask for whatever you need — like planning a stop at a restaurant or parking details — and Gemini handles the rest. Rolling out in the coming weeks on Android and iOS everywhere Gemini is…



This is all research, btw. Not just ML research. Make a hypothesis, test, examine, iterate. New knowledge is created this way.

ML research is an engineering discipline, not a philosophy seminar. You build, you test, you learn. Untested ideas are just speculation.



One thing I admire about google and deepmind is that they also go after problems that may not necessarily hold huge business value, but are good for humanity. Some things need to be done, irrespective of profit.

Together with @GoogleResearch, we’ve developed AI technologies which can monitor endangered species, protect forests, and listen to birds around the world. Here’s how. 🌱🧵



Internet became a great equalizer since 2000’s. Then social media gave everyone a voice. Today, AI is a great equalizer, giving power to a common man. I am hopeful for the future.

A guy just used @AnthropicAI Claude to turn a $195,000 hospital bill into $33,000. Not with a lawyer. Not with a hospital admin insider. With a $20/month Claude Plus subscription. He uploaded the itemized bill. Claude spotted duplicate procedure codes, illegal “double…

mukund's tweet image. A guy just used @AnthropicAI  Claude to turn a $195,000 hospital bill into $33,000.

Not with a lawyer. Not with a hospital admin insider.
With a $20/month Claude Plus subscription.

He uploaded the itemized bill. Claude spotted duplicate procedure codes, illegal “double…


Protein folding has been studied for hundreds of years. Experimental technologies such as x-ray diffraction and cryo EM were invented to get snapshots of protein structures. But nothing could solve this problem until AlphaFold 2. This is the most transformative discovery of this…

Demis Hassabis and John Jumper, the 2024 Nobel Prize laureates in chemistry, have developed an AI model to solve a 50-year-old problem: predicting proteins’ complex structures. In 2020, Hassabis and Jumper presented an AI model called AlphaFold2. With its help, they have been…

NobelPrize's tweet image. Demis Hassabis and John Jumper, the 2024 Nobel Prize laureates in chemistry, have developed an AI model to solve a 50-year-old problem: predicting proteins’ complex structures.

In 2020, Hassabis and Jumper presented an AI model called AlphaFold2. With its help, they have been…


Sometimes you just need to watch 10k videos and filter out the bad ones, to improve your video model. No hyper-parameter tuning can help.

sometimes there is no way to improve QA besides staring at the data until you become enlightened

eddybuild's tweet image. sometimes there is no way to improve QA besides staring at the data until you become enlightened
eddybuild's tweet image. sometimes there is no way to improve QA besides staring at the data until you become enlightened


No, most valuable skill would be to be able to review and debug vibe coded repos!

Vibe coding will be the single most valuable skill of 2026 People who don’t know it will start losing careers to those who do



Ok, we are already living the the future. Lets see if the product works as good as shown in the video.

NEO The Home Robot Order Today



For a very long time, @elevenlabsio was the king. Now the king has been deplatformed. There is no one winner in AI, you can just do things. Impressive model, by @cartesia_ai

We've raised $100M from Kleiner Perkins, Index Ventures, Lightspeed, and NVIDIA. Today we're introducing Sonic-3 - the state-of-the-art model for realtime conversation. What makes Sonic-3 great: - Breakthrough naturalness - laughter and full emotional range - Lightning fast -…



Saurabh Shukla reposted

We've raised $100M from Kleiner Perkins, Index Ventures, Lightspeed, and NVIDIA. Today we're introducing Sonic-3 - the state-of-the-art model for realtime conversation. What makes Sonic-3 great: - Breakthrough naturalness - laughter and full emotional range - Lightning fast -…


For researchers, celebrities have always been other researchers, from time immemorial.

when you're so deep into literature review, you start recognizing authors like they're celebrities



Just out: Detect Anything is a major leap for object detection! 🚀 The model, Rex-Omni achieves state-of-the-art zero-shot performance on benchmarks like COCO, even beating specialist models like Grounding DINO. arxiv.org/abs/2510.12798


When the world is running towards vibe coding and not writing even a single line of code, @karpathy drops something which is completely handwritten. @karpathy is always out of distribution.

Excited to release new repo: nanochat! (it's among the most unhinged I've written). Unlike my earlier similar repo nanoGPT which only covered pretraining, nanochat is a minimal, from scratch, full-stack training/inference pipeline of a simple ChatGPT clone in a single,…

karpathy's tweet image. Excited to release new repo: nanochat!
(it's among the most unhinged I've written).

Unlike my earlier similar repo nanoGPT which only covered pretraining, nanochat is a minimal, from scratch, full-stack training/inference pipeline of a simple ChatGPT clone in a single,…


United States Trends

Loading...

Something went wrong.


Something went wrong.