https://www.youtube.com/watch?v=LCEmiRjPEtQAndrej Karpathy is a computer scientist who talks about how software is changing very fast now. For a long time—about 70 years—how we wrote software did not really change much. But in the last few years, everything has changed a lot, and even changed again! This makes today a very special and interesting time to learn or work with computers.Three Generations of SoftwareAt first, there was Software 1.0. This means people wrote all the computer code themselves, using languages like C++ or Python. If you wanted the computer to do something, you had to write every step.Then came Software 2.0. In this kind, we use “neural networks.” Here, instead of writing code for every step, you give the computer lots of data and let it “learn” by itself. You don’t write the code, but you help it learn by giving examples and adjusting settings. This is called “training” the neural net.Now we have Software 3.0. This is brand new. It uses something called “Large Language Models” (LLMs), like ChatGPT. Now, you don’t even need to write code or train a neural network. You can just tell the computer what you want using normal English! The computer understands your words and creates a program from your instructions. This is a huge change, because now anyone who can write in English can “program” a computer.Before, if you wanted a computer to do something (like read a sentence and decide if it is happy or sad), you had to write code or train a neural network. But now, with LLMs, you can just ask in English and the computer does it.How Software Changed at TeslaAndrej worked at Tesla on the self-driving car program. At first, everything was written in Software 1.0, with lots of C++ code. But over time, they started using neural networks (Software 2.0), and these could do more and more. Soon, most of the old code was replaced with neural networks. The new way was better and smarter.Now, the same thing is happening again with Software 3.0. Instead of coding or training a network, you use prompts (instructions in English) and the computer does what you want. People working with computers now should know all three ways, because sometimes you need to use each one, depending on the problem.LLMs Are Like New Operating SystemsLLMs are so important that they are almost like a new “operating system” for computers. Before, we had Windows or Mac or Linux. Now, LLMs are like a new system, with some being open (free for anyone to use) and some being closed (private companies only).LLMs are very powerful, but also very expensive to make and use. Companies like OpenAI, Google (Gemini), and Anthropic spend a lot of money to build them. People use LLMs over the internet, and pay to use them, just like you pay for electricity.If an LLM stops working, people feel like the digital world has slowed down or become less smart, just like when the electricity goes off. That is how important these new tools are becoming.LLMs Act a Little Like PeopleLLMs learn by reading lots of text from the internet, so they can talk like humans. They are very smart in some ways, but also make silly mistakes sometimes. They can forget things, or give wrong answers, especially if the instructions are not clear. So, we have to be careful when using them.LLMs do not remember everything like a person does. Each time you use them, they only remember a little bit from your chat. This is called “context.” If you want them to know more, you have to remind them each time.LLMs and AppsNow, people are building new apps that use LLMs to help with things like writing code or searching for information. These apps can be partly controlled by people, so you can decide how much the computer should do and how much you want to check yourself. This is important, because sometimes the LLM can make big changes and you want to make sure everything is correct.