AI Shocks Again: New Models, Sam's Return, Mind Wipes, AI Virus and More

So in this week's AI news, we've got some crazy updates that span a broad range of topics from OpenAI welcoming back Sam Altman and introducing new board members to anthropic launching Claude 3 that outperforms Google and OpenAI's models in math and coding. We're covering Microsoft's innovative AI Explorer for Windows 11, groundbreaking security methods to mind wipe, risky AI knowledge, and a unique AI powered virus that targets email assistants. Additionally, we dive into Meta's investment in AI for video recommendations on Facebook and Instagram, a new app that diagnoses ear infections with AI, and Redfin's Ask Redfin tool that's transforming the home buying experience. Plus we're touching on Microsoft's AI Founders Club in Indonesia, the use of AI in social media interactions, advances in computer graphics and generative video models, and a significant improvement in atomic force microscopy with AI. Stick around as we delve into these developments, showing how AI continues to push the boundaries in tech, health, real estate and more. Alright so OpenAI recently announced their welcoming back Sam Altman to their board and introducing new members like Dr. Sue Desmond Hellman, Nicole Seligman and Fiji Simo. This news came after a big investigation by a law firm called WilmerHale and into why Sam Altman had to leave his CEO position before they found out that there were some trust issues between the old board and Sam, but concluded that Sam and another leader, Greg, are still the right people to lead OpenAI. During this tough time, OpenAI's employees showed a lot of support for Sam, which influenced the changes that followed. Besides bringing in new board members, OpenAI is also dealing with a lawsuit from Elon Musk, who is unhappy about how the company is working with Microsoft. Despite these challenges, OpenAI continues to grow with its ChatGPT app becoming incredibly popular and attracting huge investments. Alright now. Anthropic has recently launched its newest AI model called Claude 3. This model is a big deal because it's better at math, understanding information and coding compared to what Google and OpenAI have made. The AI model comes in three types, haiku, sonnet and Opus, with Opus being the most advanced. These versions are designed so people can choose what works best for their needs, balancing smartness, sets, speed and cost. The top model Opus, is especially noted for outshining its rivals. Amazon Web Services will include these Claude 3 models for its business customers as part of its ongoing collaboration with Anthropic. This introduction is part of the bigger competition among tech giants like Amazon, Google and Microsoft in the AI field. Meanwhile, government regulators are Watching these companies closely to ensure fair competition in the booming AI sector. Next, researchers have come up with a new way to find and remove risky information from AI systems, making them safer. This method, called mind wipe, can help stop AI from being used for bad things like cyber attacks or making bioweapons. A team from Scale AI, the center for AI Safety and other experts worked together on this project. They made a bunch of questions to see if an AI could help make dangerous weapons, then used the mind wipe method to get rid of this dangerous knowledge, which without harming the AI's other abilities too much. The idea is to make sure AI doesn't cause big problems while still working well in other areas. This approach is surely a better way to keep AI systems safe compared to older methods which were easier to get around. Even though the technique mainly focused on removing certain types of knowledge like biology and cybersecurity, it's seen as an important step in making AI safer for everyone. Now, another breakthrough. Researchers at Cornell Tech have made a computer virus that uses AI to move around on its own. This virus is a kind of worm, meaning it can spread itself without any help. The team tried this out on AI email helpers from big tech companies like OpenAI's GPT4, Google's Gemini Pro, and an open source model called Lava. They made a special input that tricks these AI models into helping the virus spread and steal personal info like names and credit card numbers. The virus can even hide in pictures, but making it easier to spread through emails without people knowing. After finding out how serious this could be, the researchers told OpenAI and Google about it. OpenAI said they're working on making their systems safer. This experiment shows us that while AI is cool for things like helping with emails, it's important to keep these systems safe from hackers and viruses. As AI becomes a bigger part of our lives, making sure it's secure is crucial. Okay, next Meta is putting a lot of money into AI technology, especially for videos on Facebook and Instagram. Tom Allison, the head of Facebook, talked about this plan at a tech conference. He said Meta is in the process of making a single AI model that would recommend videos and posts in users feeds more accurately. This AI model is part of a big plan that goes until 2026. It's not just for suggesting videos and reels anymore, but for the entire video system on Facebook. This means everything you watch could be recommended by this one AI, making things more interesting and faster to load. Before, Meta used different AI models for parts of Facebook like reels, groups and the Feed but now they want one model that can do it all, using lots of data to be smarter and more efficient. They've already seen that this new way increased how much time people spend watching reels by 8 to 10%. Besides videos, meta is also looking into using AI for digital assistance and making conversations in feeding groups better alright, next Doctors at the University of Pittsburgh have made a cool app that can tell if kids have ear infections just by looking at a video of their ear on a smartphone. This is really important because a lot of times kids get anti antibiotics they don't need, which isn't good for them. The app uses AI to check the video and can tell if there's an infection with 93% accuracy. That's way better than how doctors usually guess, which can be anywhere from 30% to 84% right. The app looks at the eardrums, shape, color and other details to make its guess. This app could help doctors a lot by making sure kids only get antibiotics when they really need them. Plus, the videos of the eardrum can be saved and shown to parents or other doctors or which helps everyone understand what's going on. The team that made this app hopes it will be used by doctors everywhere soon. Now Microsoft is adding a cool new tool to Windows 11 called AI Explorer. It's like a smart helper that can search through everything you do on your PC. You can ask it to find stuff like a list of restaurants someone mentioned or info on dinosaurs, and it will show you everything related it can find. This could be stuff on your computer or the Internet. AI Explorer can also give suggestions based on what you're looking at. For example, if you're looking at a photo, it might suggest editing it in the photos app. On March 21, Microsoft is expected to talk about some new computers called Surface Laptop 6 and Surface Pro 10. These are supposed to be really good at using Microsoft's latest AI tools. They'll be available for sale in April and June, but the AI features won't be ready until later in the year. These new features will first be for these new computers and not for the ones we already have. Everyone's waiting to see if these new tools and computers will be as good as promised, especially since people want better AI tools from Microsoft. Microsoft is also stepping up its game in Indonesia by starting the Microsoft AI Founders Club. This club is for startup founders who want to grow their businesses with AI's help. They get coaching, advice on expanding their business, tips on making a mark on LinkedIn, a sneak peek at new Microsoft AI tools and help from Microsoft engineers. They chose 18 startups from Indonesia to join this club, like mTarget and Meeting AI. MTarget uses AI to make email marketing smarter and more personal, and they did it with Microsoft's AI tech. Meeting AI turns meeting talks into text and summaries, making it easier for companies to keep track of important points. This effort is part of a bigger plan by Microsoft to support startups with cloud and AI technology, making it easier for them to innovate and make a difference in the new AI era. Now let's talk about the AI reply guys. On social media, especially on X, you've probably noticed those generic responses that don't really say much. Some are bots looking to scam, but it turns out real people are using AI for replies because they can't come up with something to say themselves. The creator of a tool called Magic Reply, Nilan Saha, says his tool helps people like CEOs and others who want to start tweeting but need a little push. It even helps people who struggle with English or are new to social media get noticed by making replies for them. Saha's tool lets you quickly create AI replies on social platforms, which some people love and others call spam. Despite the controversy, he compares Magic Reply to spell checkers, saying it still needs human approval. This makes it less likely for scammers to use, as they target a massive number of people. Looking ahead, Magic Reply plans to make posts from scratch based on a user's tweet history, but it's not the only one in the game. Everard's Pietro Scirano is testing a new AI that's good at mimicking his posting style. So we might be heading towards a future where most of what we see on social media is AI talking to AI. Alright now, yet another breakthrough in computer graphics comes from AI. Georgia Tech's Professor Boju and a team have improved simulations of things like tornadoes and underwater scenes by blending computer graphics with machine learning. This new method sets a high bar, suggesting we could simulate things we've never managed to before. Their work got a big nod in December with an award highlighting its impact. This approach, called neural flow maps, could change the game in computer graphics, much like how another innovation did for making 2D images 3D. The big challenge was making these simulations accurate. Traditional methods just couldn't compete with what AI offers. By creating a new system that uses AI smartly, the team achieved more lifelike results. They believe in using AI not to replace old methods, but to enhance them, opening new doors in computer graphics and showing the power of mixing physics with machine learning. Next up, OpenAI has wowed us again with new SORA videos showcasing museum tours and alien encounters that look almost like real movie productions. Sora, a cutting edge generative video model, has produced clips that blur the line between AI generated content and Hollywood level filmmaking. Although SORA is currently exclusive exclusive to OpenAI and a few testers, the shared videos offer a glimpse into the potential of AI in creating detailed dynamic scenes from just a single prompt. From playful dogs in the snow to intricate museum fly throughs, SORA demonstrates an impressive range of capabilities including multiple shots, effects and smooth motion in clips up to a minute long. These advancements hint at a future where generating high quality entertainment could be accessible to anyone thanks to AI models like Sora, working in tandem with sound lip syncing and production platforms pushing the boundaries of creativity and realism in digital media. Continuing on the exciting advancements There's a big leap in how we see tiny things thanks to scientists at the University of Illinois Urbana Champaign. They've upgraded atomic Force microscopy, a tool that maps out the surfaces of materials in 3D. This upgrade lets AFM see details even smaller than its probe tip, a game changer for studying materials and developing nanotechnology. Usually AFM's ability to see tiny things is limited by how big the probe tip is, but with this new AI method, that's not a problem anymore. The AI can ignore the probe size and show us clearer three dimensional images of material surfaces, making it much better at finding tiny features. To do this, the AI breaks down the images to get past the blur caused by the probe tip and reconstructs them them to show the real surface details. This breakthrough means we can now get a closer, more precise look at materials, which is crucial for making advancements in electronics and understanding materials better. This development is just the start, promising even more precise material studies with better AI training in the future. Alright now Redfin, a big name in real estate, is introducing a smart helper called Ask Redfin. This AI tool can quickly give you answers about houses you're interested in Redfin. It uses smart technology and Redfin's own data to provide info, and if it gets stuck, it connects with Redfin's support team to help out. This tool is designed to make it easier for people to find what they need without having to dig through a lot of information. However, there's been some concern about fairness. Last year there was worry that AI tools in real estate could be biased. Redfin says it's tested Ask Redfin to make sure it doesn't break any housing laws or be unfair. For now, Ask Redfin is being tested in some places in the US and you can try it out in the Redfin app. Alright, that concludes this week's AI News. If you enjoyed this video, don't forget to hit the like button, subscribe to my channel and click the Bell icon to stay updated on all my future uploads. Thanks for tuning in and I'll catch you in the next one.