Get live statistics and analysis of Keerthana Gopalakrishnan's profile on X / Twitter

It’s called a white chikankari kurta. The chikankari embroidery style has its origins in Lucknow, India but is now staple fashion worn across the country.

A 15 story building came back from space and they caught it with a pair of tongs attached to a tower?? If you work with hardware you *know* this represents the best of engineering! Mind-blowing and inspiring!

They don’t give Nobels for engineering feats and that’s a mistake. Elon and the folks at SpaceX deserve one for making reusable rockets possible, and working steadfastedly on it for over two decades! This changes the trajectory of humanity than many other inventions.

Gemini Nano runs on a phone and without the internet, beginning an era of on-device LLMs, one that fits in your pocket

Gemini Robotics is out and is the most advanced VLA in the world! I am especially blown away by the instruction following results: it is the first time where I personally have felt that building generic embodied intelligence is within reach! Like a robot coming to life.

In the last two years, large foundation models have proven capable of perceiving and reasoning about the world around us unlocking a key possibility for scaling robotics. We introduce a AutoRT, a framework for orchestrating robotic agents in the wild using foundation models!

Solve AGI, use it to solve everything else, including cancer 🙃

1X doesn’t deserve half the hate they’re getting! As someone who works on humanoids, I’m incredibly impressed by their bold moves: making a humanoid that’s safe enough to be in homes, and building a large data flywheel in the wild, is an interesting approach and one that is crazy enough that it may just work. In the age of data driven learning, deployment precedes automation. We don’t need perfect hardware to build a useful robot. There’s a lot of intelligence frontier to be captured on a general purpose robot that can do lots of useful things even if it is not at the forefront of dexterity. As a community, let us be open minded to encourage idea diversity and taking risk.

All serious robotics companies are foundation model companies. Acting intelligently in the physical space is an emergent property of a large audio visual language agent, and intelligence at the most fundamental level is the same whether its expression is digital or physical.

For folks affected by the Meta layoffs, Deepmind is hiring! Could not be more upbeat about the future :)

NeurIPS was wild. 20000+ people, so much excitement, really feels like we are living through a critical point in human history. Also met with the cutest RL researcher in the hallways:

Are 10M context models the beginning of the end for RAG? Why retrieval when you’ve near infinite memory?

Prompt engineering will not be a thing in the future. It’s hacky af and will be dead as soon as models become smarter in understanding what you’re asking for. That will also be the hill on which prompt engineer as a job and prompt tuning/ prompt markets start ups will die on.

Are 10M context models the beginning of the end for RAG? Why retrieval when you’ve near infinite memory?

They don’t give Nobels for engineering feats and that’s a mistake. Elon and the folks at SpaceX deserve one for making reusable rockets possible, and working steadfastedly on it for over two decades! This changes the trajectory of humanity than many other inventions.

It’s been like 2 months since Claude code and while everyone’s talking about being 10x-ed and writing xxxx more likes of code, I’m yet to see a lot of impactful passion projects. May be our world really is ideas bottlenecked?

Gemini Nano runs on a phone and without the internet, beginning an era of on-device LLMs, one that fits in your pocket

What are some books on philosophy that influenced you?

1X doesn’t deserve half the hate they’re getting! As someone who works on humanoids, I’m incredibly impressed by their bold moves: making a humanoid that’s safe enough to be in homes, and building a large data flywheel in the wild, is an interesting approach and one that is crazy enough that it may just work. In the age of data driven learning, deployment precedes automation. We don’t need perfect hardware to build a useful robot. There’s a lot of intelligence frontier to be captured on a general purpose robot that can do lots of useful things even if it is not at the forefront of dexterity. As a community, let us be open minded to encourage idea diversity and taking risk.

If you’ve published in robotics / AI before and have some time over the next couple days, we’d love to enlist your service as an emergency reviewer for CoRL 2024! Topics: VLMs, diffusion, RL Comment below if you’re available and I’ll DM you :)

Gemini Robotics is out and is the most advanced VLA in the world! I am especially blown away by the instruction following results: it is the first time where I personally have felt that building generic embodied intelligence is within reach! Like a robot coming to life.

Prompt engineering will not be a thing in the future. It’s hacky af and will be dead as soon as models become smarter in understanding what you’re asking for. That will also be the hill on which prompt engineer as a job and prompt tuning/ prompt markets start ups will die on.

Solve AGI, use it to solve everything else, including cancer 🙃

India should build its own foundation models

Ok I’ll bite. Is Apple buying 1x? Is that the news of Oct 28? Can we get a prediction market for this?

In the last two years, large foundation models have proven capable of perceiving and reasoning about the world around us unlocking a key possibility for scaling robotics. We introduce a AutoRT, a framework for orchestrating robotic agents in the wild using foundation models!