Google's DeepMind New AI system - AlphaGeometry x Sam Altman's Billion-Dollar Chip Factory

Sam Altman, the CEO of OpenAI, is on a mission to raise a huge amount of money. Aiming for billions of dollars, he's got a big plan to start a company that makes semiconductors which are essential for powering AI technologies. To make this happen, he wants to build a whole network of factories where these chips can be manufactured. This is no small task. It's going to need a lot of money and several years to set up. So recently at the World Economic Forum in Davos, Switzerland, which took place from January 15 to 19, 2024, Altman discuss this ambitious plan. Bloomberg has reported that Altman is currently in talks with various potential investors to gather the billions of dollars needed for this chipmaking venture. Altman, who is 38 years old, is not just thinking small. He wants to establish a whole network of factories that can produce these semiconductors. Setting up just one state of the art manufacturing facility for these chips is really expensive. We're talking tens of billions of dollars and Altman doesn't want to stop at just one. He's planning to create several of these facilities which is a project that will take a lot of time and even more money. Among the big names Altman has approached for funding is Sheikh Tanun Bin Zayed Al nahyan from the UAE, the Taiwan Semiconductor Manufacturing Company and SoftBank. Sheikh Tanun is not just any investor. He's the brother of the UAE President, Sheikh Mohammed Bin Zayed Al Nahyan and also works as the National Security Advisor. He's in charge of some of the UAE's most influential investment funds. According to the Financial Times, there's still a question about whether this new chipmaking company will be a part of OpenAI or a completely separate entity. However, insiders say that OpenAI is expected to be the first customer of this new venture. Reuters reported last year that OpenAI was already thinking about making its own chips and was even considering buying a company that could help with this. While nothing was set in stone, the company was exploring different ways to deal with the shortage of AI chips. They were also looking at working with other chip makers aside from their current supplier Nvidia. So why are these AI chips so important? Well, ever since OpenAI released ChatGPT over a year ago, there's been a huge surge in interest in AI from both regular people and businesses. This has led to a big increase in demand for the processors that run AI programs. That's where Sam Altman's big plan for a chipmaking venture comes in. Alright, now Google's DeepMind team has built a new AI system called Alpha Geometry that's really good at solving tough geometry problems. It's a big deal because it shows AI getting closer to thinking like humans, especially in tricky areas like math. Geometry has always been a tough nut to crack for AI. Unlike text based AI, there's not much training data for math, since it's all about symbols and specific domains. Feng Luong, who helped write the research, says that this is why math is a great way to see how smart AI is getting. Most AIs aren't great at logical reasoning, which you really need for math. But alpha geometry is different. It mixes a language model, which is good at spotting patterns and guessing next steps, with a symbolic engine, an AI that uses symbols and rules to figure things out. Language models are good at some parts of thinking, but not so much at the strict reasoning needed for math. The symbolic engine fills in this gap using strict rules and logic. These two parts, one for creative thinking and the other for logical reasoning, work together to tackle hard math problems. This is a lot like how we humans do geometry, using what we know and experimenting to find answers. Deep Mind tested alpha geometry with geometry problems from the International Mathematical Olympiad, where the smartest high school students compete. It solved 25 out of 30 problems, which is way better than the previous best system made by Chinese mathematician wensun Wu in 1978, which only solved 10. Floris Van Dorn, a math professor not involved in the research, was really impressed. He thought this kind of progress was still years away. DeepMind believes alpha geometry shows how AI can reason and even discover new math knowledge. Kwak v. Lee, a DeepMind scientist, says this is another sign of how AI can help us understand the world better. When alpha geometry faces a geometry problem, it first tries to solve it using its symbolic engine. If that doesn't work, the language model steps in and adds something new, like a point or line. This gives the symbolic engine more to work with. This back and forth continues until they find a solution. To train the language model, the team had to create a lot of data themselves because there wasn't enough existing geometric data. They made nearly half a billion random geometric diagrams, and the symbolic engine analyzed each one to produce statements about their properties. They used these statements to make 100 million synthetic proofs for training. Roman Yampolski, a computer science professor not involved in the research, thinks alpha geometry skills are a big step toward more sophisticated human like problem solving in machines. He says it could be useful in fields that need geometric problem solving, like computer vision, architecture and theoretical physics. But there's still room to grow Alpha geometry is good at elementary math, but not yet at the more abstract stuff taught at universities. Van Dorn thinks mathematicians would be really interested if AI could solve research research level math problems, maybe by having new mathematical insights. Luong says the plan is to apply this approach to more areas of math. Geometry is just the start, showing that AI is close to being able to do deep reasoning now. Google also just announced some cool new features for Android Auto powered by AI to make driving less distracting. This big news comes right when Samsung's launching their Galaxy S24 series. So Android Auto is Google's version of Apple CarPlay. It's all about making your phone's car friendly stuff easy to use on your car's display. But if you've ever tried to reply to a text while driving, you know it's not always smooth sailing. Dealing with a bunch of messages or tricky voice assistants can be as distracting as actually looking at your phone. Google's got a plan to fix this with their new Android Auto updates. The idea is to use AI to give you quick summaries of long texts or busy group chats while you're driving. This way you're not tempted to look at your phone. This smart tech will also suggest actions and replies and do them for you so you don't even need to touch your phone. Here's how it say a friend texts you about lunch. Android Auto will pick out the important details like where you're eating and when. It can then send your friend your arrival time or even call them. All you do is tap once to get directions to the restaurant and you're set. This seems like a smart way to use AI. Being able to change your route or send an ETA with just a click means you can focus more on driving and less on your car's display. Plus, having fewer distracted drivers is definitely a good thing. Alright, that wraps up the video. I hope you found it as interesting as I did. If you did, smash that like button. And don't forget to subscribe to the channel for more AI content. Thanks for watching and see you in the next one.