AI STUNS AGAIN: Microsoft Copilot's New Features and Quiet-STaR's Breakthrough!

So Microsoft just introduced two powerful new computers with advanced AI features called the Surface Pro 10 and the Surface Laptop 6, specially designed for businesses and come with some cool AI tools that weren't available before. Like one neat thing is a special button on the keyboard that lets you use the AI helper instantly. In this video, we are going to talk about these devices and the latest updates to 2Copilot. Later on we'll dive into an exciting new AI model for autonomous materials science. Plus, we'll look into a groundbreaking AI training technique that sharpens reasoning skills by promoting an inner monologue. Make sure to stick around because all these topics are super interesting. Let's kick things off with the latest from Microsoft. This move by Microsoft is a strong indication that artificial intelligence is becoming an integral part of the computing experience experience, especially in the professional world. It suggests that AI might be more than just a fleeting trend, hinting at its potential to become a permanent fixture in technology. The cool thing about these new features is how AI is mixed into a bunch of Microsoft products like Windows, Microsoft 365 Teams and Edge. The goal here is to make using AI feel smooth and natural across everything you do on your computer. This way, it helps you get more done and makes your tasks easier easier without you even noticing it's there. For example, dealing with a flooded inbox after a vacation can be daunting. However, The AI in Microsoft 365 can simplify this task by highlighting the most urgent emails and summarizing the content that matters most to you. Similarly, if you missed a meeting, the AI can provide you with a concise summary of the discussions, decisions made, and action items, ensuring you're caught up without having to sift through minutes or recordings. Writing emails can also be streamlined with AI's assistance. If you need to update your team about a meeting, the AI can draft the email for you, tailoring it to your instructions. This means if you want the email to be concise and targeted at a senior executive audience, the AI will adjust its tone and content accordingly. This feature not only saves time, but also ensures that the message is appropriate for its intended recipients. Beyond administrative tasks, AI extends its utility to web research and content creation. This capability is particularly particularly useful for those who rely on up to date information and creative content generation in their daily tasks. It can also alleviate the need for IT support for routine issues and settings adjustments on your computer. From changing your wallpaper to enabling Do Not Disturb mode or managing power settings, AI can handle a variety of tasks that would otherwise require navigating through menus and settings. This not only enhances user convenience, but but also reduces the workload on IT departments. In the realm of security, the emphasis has grown, especially with the increase in remote and hybrid work models. The new devices introduced by Microsoft are equipped with advanced security features to protect sensitive information. These include measures like single sign on and encryption, along with tools designed to prevent data leaks such as restrictions on screen captures and watermarking. Additionally, management tools for software and compliance are part of the package highlighting Microsoft's commitment to secure and efficient operations. Microsoft's vision with these AI tools is to guide businesses towards an AI centric approach. While the benefits of AI are clear in terms of efficiency and productivity, the technology also faces scrutiny over issues like privacy, environmental impact and potential job displacement. Alright, now moving from the office to the laboratory, let's explore Explore how AI is also revolutionizing the field of material science at the Pacific Northwest National Laboratory, scientists have created a new AI model that's really good at finding patterns in pictures taken by electron microscopes of different materials all by itself, without needing people to help it. This is big news for materials science, a field that helps us make everything from lightweight cars and strong computers to batteries that can hold a lot of charge, and spacecraft that are very durable. Normally, looking at materials to understand them better takes a lot of effort and time because you have to look through so many tiny details. Stephen Spurgeon, a senior materials scientist at pnnl, and his team have been working for a long time to use AI to help in material science. They study new materials for things like catalysts, energy storage and electronics. They also look into how materials change over time when they are used in places like nuclear reactors or spacecraft, where there's a lot of radiation that can damage the materials. Oh my God. Usually, to get an AI model to recognize something like radiation damage in materials, researchers would have to manually mark the damaged areas in pictures from the electron microscope. This manual process is slow and sometimes not very accurate because people can make mistakes or be inconsistent in what they consider damaged. So the team at PNNL came up with a new approach using a an unsupervised model. This means the model can figure things out on its own without needing humans to label the pictures first. They started with an existing AI model called ResNet50 and a huge collection of over 100,000 electron microscopy images that weren't labeled, called Micronet. They trained the model to look at the pictures in small sections, compare them, and group similar sections together into what they call communities. This way, the model can show patterns in the images by grouping similar areas together, all without being told what to specifically look for. This AI model has been used to understand how materials get damaged by radiation. It's really good at finding and grouping damaged areas in the images, which helps scientists understand the materials better. Armentera Petrosian, a research associate at pnnl, explained that this method lets them see how different areas relate to each other, even if they're not close in the material. The great thing about this model is that it's very consistent in finding these patterns, more so than humans might be. This consistency is super helpful, not just for looking at images, but also for coming up with clips, clear ways to describe how materials change under different conditions. Electron microscopes don't just take pictures. They can also gather other kinds of data like spectroscopy readings and diffraction patterns. With human labeling, researchers usually could only work with one type of data at a time. But this new AI model can handle multiple types of data together, making it more powerful and able to predict things better. This AI model is part of moving towards fully automated experiments with electron microscopes. And at pnnl, they already have a project called Autoem that uses AI to combine and identify features in electron microscope images, quickly helping researchers focus on the most interesting parts. This new model takes it further by quickly finding and grouping similar areas, speeding up the research process. Now the team is working on making the model faster and able to understand more types of data and more complex situations. They're also focusing on using the model in real time so it can help with experiments as they're happening. This could set an example for how other researchers can use similar models in their work. Alright, now here's another exciting development in AI. Scientists have developed an innovative method to enhance AI intelligence. Known as Quiet Start, it helps AI think before it talks, kind of like how we do. They tell the AI to come up with different ideas inside its head before giving an answer. Finally, someone who speaks English. Venki Yarapodu from 4 Kris says this will make AI much better because it'll understand people more like how we understand each other, catching all the little hints in the way we mean things. The big deal with Quietstar is that it's different from usual AI chatbots like ChatGPT. Instead of just spitting out replies, it thinks about what could come next in a conversation and learns as it goes. This new method got tested on Mistral 7B and it did way better on a reasoning test after training with quietstar, even though it still found a school math test tough. But hey, it got better at that too. Eric Zelickman shared some exciting news about how this training not only makes the AI smarter in one way, but boosts its smarts across the board. It gets better at understanding common sense and solving math problems it hasn't seen before. Why does AI need to think like us? Well, we humans can quickly get the gist of simple stuff like math because of our experiences. AI is trying to get the hang of this, but it's not perfect. Benny Gill from Cognitos points out that we humans can only focus on a few things at a time, and AI has the same issue. That's why when AI tries to think step by step or have an inner monologue, it's mimicking us, trying to sort things out in our heads or on paper. Quietstar isn't the only smart way to teach AI to think like us. Microsoft came up with something called Algorithm of Thoughts last year, guiding AI to solve problems in a way that's more like how we do it. Making AI think like humans is also about making it easier for us to understand how AI makes decisions. There's this cool method called deep distilling developed by researchers which is inspired by how our brains work. It simplifies complicated info into hubs, kind of like breaking down a big idea into smaller, easier to get pieces. This human like thinking isn't just for show, it could make robots much smarter too. Covariance New Robotics foundation model RFM1 is all about giving robots the ability to understand language and the world around them better. Yarapotu believes that AI will keep getting better at figuring things out on its own, understanding natural language and solving problems with less help from us. So it looks like AI is on its way to becoming more and more like a smart friend who gets us alright. Don't forget to hit that subscribe button for more updates and subscribe. Thanks for tuning in and we'll catch you in the next one.