Google Gemini AI is Launching Now...
Skip the waitlist and invest in blue-chip art for the very first time by signing up for Masterworks: https://www.masterworks.art/airevolution
Purchase shares in great masterpieces from artists like Pablo Picasso, Banksy, Andy Warhol, and more.
See important Masterworks disclosures: https://www.masterworks.com/cd

Google DeepMind introduces "Gemini", aiming to create a universal AI that can process and generate various content types like text, images, and videos. Meanwhile, MIT and Harvard unveil "FAn", using Vision Transformers for real-time robot object tracking. Both advancements highlight breakthroughs in AI, offering promising futures for tech enthusiasts and businesses.

Become a Member of the channel and Supporter of AI Revolution → https://www.youtube.com/channel/UC5l7RouTQ60oUjLjt1Nh-UQ/join

#Gemini #Google

As we all know, Google recently launched a new AI assistant, which many believe is a trial run for an upcoming project named Gemini.

They're trying out various AI features to see how users respond.

Gemini is anticipated to integrate everything from AlphaGo to Google's AI search.

It aims to be the most powerful AI system ever made, potentially transforming the internet and our daily lives.

In this video, we'll discuss the Gemini project.

Later on, we'll also cover a new AI project from MIT and Harvard University called FAN.

This is another groundbreaking development, so it's worth watching the video till the end to learn all about it.

So initially, Gemini was a product of the Gemini Project by Google DeepMind, the group behind AlphaGo, the AI that defeated the Go World Champion in 2016.

The aim of the Gemini project is to build a universal AI that can tackle any task with any kind of data without specific models.

Gemini is the initial phase of this project.

It's a big language model that processes text, images, videos, and more.

It can even create content, like turning text into a video or turning speech into an image.

Gemini uses techniques from AlphaGo, including reinforcement learning, training AI through feedback, and TreeSearch, exploring possible action outcomes.

By mixing these with language models, Gemini can address challenges in various areas.

What's special about Gemini is its architecture that focuses on handling different data types simultaneously.

For instance, if you provide a text describing a scene, Gemini can create a corresponding image, video, and sound.

Conversely, From an image, video, or sound, it can produce a descriptive text.

Gemini has an advantage over other AI systems because it can handle multiple types of content, like text, images, and audio, all at once.

In contrast, OpenAI's chat GPT is great at creating text, but struggles with images, videos, or audio.

If you wanted to use OpenAI for those, you'd have to use different models, like DAL-E for pictures or Jukebox for songs.

First, Google sees potential in improving their current tools and products with Gemini.

For example, their chatbot, BARD, and their search engine could benefit.

Think about asking Gemini anything and getting an answer in any format you like.

It's efficient and can quickly solve problems using Google's vast resources.

Second, Google has a lot of data, more than many of its rivals.

This data comes from places like YouTube, Google Books, their main search index, and academic content from Google Scholar.

Using all this information, Google can train better models and produce varied and innovative results.

And third, Google plans to offer Gemini to users of its cloud platform.

This means businesses and developers could use Gemini's abilities for their projects.

They might develop unique learning resources, create assistive tech, or generate new content using ambient computing.

So when can we expect to see Gemini in action?

Well, Google hasn't announced an official release date yet, but they have said that they will reveal more details about the project in the fall of this year.

So stay tuned for more updates on this exciting development.

In the meantime, let me know what you think about Gemini in the comments below.

Do you think it will surpass ChatGPT and other AI systems?

What kind of content would you like to see Gemini generate?

How would you use Gemini if you had access to it?

It's still crazy to see how quickly this all escalated.

I mean, before most people even started using chat GPT, AI played a big role in boosting the US economy.

While this is great news for many of us, it's worrying that aside from big tech companies using AI, many other companies aren't doing well financially.

Last year, some top investors started buying assets like fine art instead of stocks to spread their risks.

Today's sponsor, Masterworks, offers you this diversification strategy, which was once exclusive to the wealthiest people on earth.

They've compiled decades worth of auction data to invest in art they believe will appreciate in value.

They buy it up front, qualify it with the SEC, and break it into investable shares.

Net proceeds from its eventual sale are then distributed to its investors.

Now, I'm not a financial advisor, and historical returns are not a guarantee for future returns, but the results are so impressive.

In just a few years, they've sold over $45 million worth of artwork.

And just a couple of weeks ago, they sold a Cecily Brown painting for a jaw-dropping 77% annualized net returns, bringing their streak to 15 straight profitable exits.

Masterworks has over 800, 000 users, and their art offerings have sold out within hours, which is why they've had to make a waitlist.

But my viewers can skip the line and get priority access right now by clicking the link in the description.

Alright now, let's discuss FAN, short for Follow Anything.

This is a new system developed by MIT and Harvard researchers that allows robots to track any object in real time using just a camera and a simple query, whether it's text, image, or a click.

In this video, I'll break down what FAN is and why it's impressive.

FAN uses the transformer architecture for visual object tracking.

Transformers, commonly known for advancing natural language processing, NLP, can do things like generate text and translate languages.

The researchers were curious to see if transformers could also be effective with images.

You see, most of the existing robotic systems that can follow objects use convolutional neural networks, or CNNs.

These are another type of neural network that can process images by applying filters and pooling operations.

CNNs are great for tasks like image classification and segmentation, but they have some limitations when it comes to tracking and following objects.

For example, they can only handle a fixed set of object categories that they have been trained on.

They also require a lot of manual tuning and calibration to work well in different environments and scenarios.

And they are not very user-friendly because they often need complex inputs like bounding boxes or masks to specify the target object.

Phan solves these problems by using a different approach.

Instead of CNNs, It uses vision transformers or VITs.

These are transformers that can process images by splitting them into patches and treating them as sequences of tokens.

VITs can learn to capture the relationships between different parts of an image, just like Transformers can capture the relationships between different words in a text.

And because they are based on attention mechanisms, they can focus on the most relevant parts of the image for the task at hand.

FAN uses VITs for real-time tracking and segmentation of objects in videos.

It identifies the object and distinguishes it from the background.

To get started, all it requires is a bounding box.

After that, you can guide Fan to recognize new objects by typing a description, showing a picture, or clicking on the object in the video.

For instance, if you want Fan to track a red ball, type red ball, show a picture of 1, or click on it in the video.

Fan will then track the red ball throughout the video.

You can easily switch to a different object by changing your instruction.

What's impressive is Fan isn't just limited to tracking 1 item.

It can track multiple objects simultaneously by just giving separate instructions for each.

Phan has shown impressive performance in visual object tracking and segmentation, achieving top results in real time.

It operates at about 55 frames per second on a standard GPU and can tackle challenges like occlusions, fast motion, and background disturbances.

When compared to popular CNN-based methods like Siam Mask and SegaRot, FAN was more accurate and robust.

Unlike those methods, FAN can work across various data sets without extra training.

This progress suggests a future where robots can easily and smartly interact with any object in any setting.

Imagine a robot assistant that understands your commands and does tasks like fetching or cleaning.

Or a robot that can play games and explore unknown places, so the future looks promising, I guess.

And the best part is that Phan is not some proprietary technology that only a few people can access.

The researchers have made their code and models available online for anyone to use and improve.

You can find it on GitHub repository and I highly encourage you to check it out and try it for yourself.

If you did, make sure to hit the like button and subscribe to my channel.

And remember to click the bell icon to stay updated on new uploads.


Speaker 0: As we all know, Google recently launched a new AI assistant, which many believe is a trial run for an upcoming project named Gemini. They're trying out various AI features to see how users respond. Gemini is anticipated to integrate everything from AlphaGo to Google's AI search. It aims to be the most powerful AI system ever made, potentially transforming the internet and our daily lives. In this video, we'll discuss the Gemini project.

Later on, we'll also cover a new AI project from MIT and Harvard University called FAN. This is another groundbreaking development, so it's worth watching the video till the end to learn all about it. So initially, Gemini was a product of the Gemini Project by Google DeepMind, the group behind AlphaGo, the AI that defeated the Go World Champion in 2016. The aim of the Gemini project is to build a universal AI that can tackle any task with any kind of data without specific models. Gemini is the initial phase of this project.

It's a big language model that processes text, images, videos, and more. It can even create content, like turning text into a video or turning speech into an image. The potential uses are vast. Gemini uses techniques from AlphaGo, including reinforcement learning, training AI through feedback, and TreeSearch, exploring possible action outcomes. By mixing these with language models, Gemini can address challenges in various areas.

What's special about Gemini is its architecture that focuses on handling different data types simultaneously. For instance, if you provide a text describing a scene, Gemini can create a corresponding image, video, and sound. Conversely, From an image, video, or sound, it can produce a descriptive text. Gemini has an advantage over other AI systems because it can handle multiple types of content, like text, images, and audio, all at once. In contrast, OpenAI's chat GPT is great at creating text, but struggles with images, videos, or audio.

If you wanted to use OpenAI for those, you'd have to use different models, like DAL-E for pictures or Jukebox for songs. With Gemini, it's all combined. So why is Google working on Gemini? There are a few reasons. First, Google sees potential in improving their current tools and products with Gemini.

For example, their chatbot, BARD, and their search engine could benefit. Think about asking Gemini anything and getting an answer in any format you like. It's efficient and can quickly solve problems using Google's vast resources. Second, Google has a lot of data, more than many of its rivals. This data comes from places like YouTube, Google Books, their main search index, and academic content from Google Scholar.

Using all this information, Google can train better models and produce varied and innovative results. But we need to go. And third, Google plans to offer Gemini to users of its cloud platform. This means businesses and developers could use Gemini's abilities for their projects. They might develop unique learning resources, create assistive tech, or generate new content using ambient computing.

So when can we expect to see Gemini in action? Well, Google hasn't announced an official release date yet, but they have said that they will reveal more details about the project in the fall of this year. So stay tuned for more updates on this exciting development. In the meantime, let me know what you think about Gemini in the comments below. Do you think it will surpass ChatGPT and other AI systems?

What kind of content would you like to see Gemini generate? How would you use Gemini if you had access to it? It's still crazy to see how quickly this all escalated. I mean, before most people even started using chat GPT, AI played a big role in boosting the US economy. While this is great news for many of us, it's worrying that aside from big tech companies using AI, many other companies aren't doing well financially.

Last year, some top investors started buying assets like fine art instead of stocks to spread their risks. Today's sponsor, Masterworks, offers you this diversification strategy, which was once exclusive to the wealthiest people on earth. They've compiled decades worth of auction data to invest in art they believe will appreciate in value. They buy it up front, qualify it with the SEC, and break it into investable shares. Net proceeds from its eventual sale are then distributed to its investors.

Now, I'm not a financial advisor, and historical returns are not a guarantee for future returns, but the results are so impressive. In just a few years, they've sold over $45 million worth of artwork. And just a couple of weeks ago, they sold a Cecily Brown painting for a jaw-dropping 77% annualized net returns, bringing their streak to 15 straight profitable exits. Masterworks has over 800, 000 users, and their art offerings have sold out within hours, which is why they've had to make a waitlist. But my viewers can skip the line and get priority access right now by clicking the link in the description.

Alright now, let's discuss FAN, short for Follow Anything. This is a new system developed by MIT and Harvard researchers that allows robots to track any object in real time using just a camera and a simple query, whether it's text, image, or a click. In this video, I'll break down what FAN is and why it's impressive. FAN uses the transformer architecture for visual object tracking. Transformers, commonly known for advancing natural language processing, NLP, can do things like generate text and translate languages.

The researchers were curious to see if transformers could also be effective with images. You see, most of the existing robotic systems that can follow objects use convolutional neural networks, or CNNs. These are another type of neural network that can process images by applying filters and pooling operations. CNNs are great for tasks like image classification and segmentation, but they have some limitations when it comes to tracking and following objects. For example, they can only handle a fixed set of object categories that they have been trained on.

They also require a lot of manual tuning and calibration to work well in different environments and scenarios. And they are not very user-friendly because they often need complex inputs like bounding boxes or masks to specify the target object. Phan solves these problems by using a different approach. Instead of CNNs, It uses vision transformers or VITs. These are transformers that can process images by splitting them into patches and treating them as sequences of tokens.

VITs can learn to capture the relationships between different parts of an image, just like Transformers can capture the relationships between different words in a text. And because they are based on attention mechanisms, they can focus on the most relevant parts of the image for the task at hand. FAN uses VITs for real-time tracking and segmentation of objects in videos. It identifies the object and distinguishes it from the background. To get started, all it requires is a bounding box.

After that, you can guide Fan to recognize new objects by typing a description, showing a picture, or clicking on the object in the video. For instance, if you want Fan to track a red ball, type red ball, show a picture of 1, or click on it in the video. Fan will then track the red ball throughout the video. You can easily switch to a different object by changing your instruction. What's impressive is Fan isn't just limited to tracking 1 item.

It can track multiple objects simultaneously by just giving separate instructions for each. Phan has shown impressive performance in visual object tracking and segmentation, achieving top results in real time. It operates at about 55 frames per second on a standard GPU and can tackle challenges like occlusions, fast motion, and background disturbances. When compared to popular CNN-based methods like Siam Mask and SegaRot, FAN was more accurate and robust. Unlike those methods, FAN can work across various data sets without extra training.

This progress suggests a future where robots can easily and smartly interact with any object in any setting. Imagine a robot assistant that understands your commands and does tasks like fetching or cleaning. Or a robot that can play games and explore unknown places, so the future looks promising, I guess. And the best part is that Phan is not some proprietary technology that only a few people can access. The researchers have made their code and models available online for anyone to use and improve.

You can find it on GitHub repository and I highly encourage you to check it out and try it for yourself. Alright, thanks for watching the video. I hope you found it informative. If you did, make sure to hit the like button and subscribe to my channel. And remember to click the bell icon to stay updated on new uploads.

Thanks again and I'll see You

See all AI Revolution transcripts on Youtube

Google Gemini AI is Launching Now + FAn: Real-time Robotic Vision Breakthrough