Which of these is the most powerful AI company? Is it A. OpenAI, B. Google, C. Meta or D.
None of these? Many people argue that it's none of these. All of these companies have done a great job at building foundational AI models.
But they rely on NVIDIA's CPUs to train them. None of this generative AI magic would be possible if NVIDIA did not exist. Today I want to show you how you can learn AI from NVIDIA and get certified by them.
Let's do this. NVIDIA offers many courses to learn the essentials of AI. Most of these courses are free, but there are some that are paid as well. Let's start with some free courses. Let's cover the basics first.
The first course I want to talk about requires no programming knowledge. The course is called Generative AI Explained. Generative AI is a type of artificial intelligence that can create new content like images, text, or music. Think of tools like DALL-E creating images from text descriptions or ChatGPT generating human-like responses.
The course will define Generative AI and explain how it works. It will show you all the different applications of Generative AI and tell you about the potential opportunities in the field. You will learn about the underlying technologies behind Generative AI such as Neural Networks and Deep Learning.
By the end of this course, you will have a good understanding of what Generative AI is capable of and how it is shaping different industries. Moving on, if you have ever worked with GPT or any other similar large language model, you would know that sometimes it can just make stuff up. This is also called hallucinations.
For this reason, most people still use Google to find information that is more factually correct. But if you are following this space closely, you would know that there are startups like Perplexity that provide chat GPT like text generation but are factually accurate. So how do they do it? They use something called Retrieval Augmented Generation or RAG. Let's break down the difference between GPT and RAG.
down what retrieval augmented generation means. Retrieval simply means finding relevant information from a knowledge base. This could be a database of facts, articles or search results. Augmented means improving something.
In this context, it means improving AI's knowledge with the help of retrieved information. Generation refers to creating new content which is what the large language model does. In simple words, with drag, we retrieve the relevant information and use it to augment AI's knowledge before generating a response. It's an end-to-end architecture that combines the information retrieval component with a response generator. This approach helps to ground the AI's responses in factual information, reducing the likelihood of hallucinations.
If you want to learn more about how to use RAG with LLMs, you can do this course called Augment Your LLM Using Retrieval Augmented Generation by NVIDIA. This course will cover the RAG components that NVIDIA uses internally. It's a great way to start your journey with LLMs and RAG. In the beginning of this video, we talked about how most companies are using NVIDIA's CPUs to train models. But what makes NVIDIA's CPUs so special?
It's their parallel processing power. NVIDIA's CPUs contain thousands of smaller cores that can perform tasks simultaneously. This makes them a go-to choice for AI model training where parallel processing is critical.
Let's understand why that is. Consider the task of training a large language model. These models usually have billions of parameters that need to be adjusted during the training. If you were to do it sequentially, one calculation at a time, it would take an impractically long amount of time to train. But with parallel processing, you can perform these calculations simultaneously.
It significantly reduces the time to train. Compute Unified Device Architecture or CUDA allows developers to tap into GPU's parallel processing power. In simple words, CUDA makes it easy to write code that can be run in parallel on a GPU.
If you want to learn how to write parallel CUDA kernels, you can do this course called an even easier introduction to CUDA. This course will teach you how to organize parallel thread execution and efficiently manage memory. If you have played around with chat GPT or cloud, you would know that the response you get from them is as good as the questions you ask them.
The art of crafting effective prompts is becoming increasingly important as AI models become more sophisticated and widely used. Asking the right questions or prompt engineering is going to become a highly valuable skill in the future. To learn prompt engineering, you can do this course called Prompt Engineering with LAMA 2. For those who don't know, LAMA is the open source last language model from Meta. This course will teach you how to write precise prompts to achieve the desired LLM behavior.
You will use techniques like few-shot learning where you provide examples in your prompt to guide model's output. We have talked a lot about Generative AI, LLMs and ChatGPT so far. All these are applications of something called Deep Learning.
Deep Learning is a subset of machine learning techniques where we use artificial neural networks to learn from large amounts of data. You can get started with Deep Learning by using this course. All the courses that we have talked about before this one are less than 3 hours and are more high level. This course is much more hands on.
This course goes a little deeper. You would need to know Python if you want to do this course. This course will teach you PyTorch which is a popular deep learning library.
You will also learn Convolutional Neural Networks or CNNs. CNNs are particularly effective for processing grid-like data like images. This course also covers Transfer Learning and Natural Language Processing. All the courses we discussed can be done separately. But if you like slightly more structured approach, Nvidia also offers some learning paths.
They have a foundational learning path for basics. This path is for those who are new to AI and want to build a solid foundation. They also have a path for generative AI and LLMs. This path goes deeper into the technology-powering latest AI breakthroughs. It covers topics like transformer architecture and diffusion models.
So far we have only talked about how to learn new AI skills. But what if you want to get certified for your skills? NVIDIA also offers certification for different skills. These certifications can be a great way to validate your skills and stand out in the job market.
For example, you can become NVIDIA certified for generative AI LLMs after doing some of the courses we discussed today. They also offer some other certificates like AI Infra and Operations Certificate and Multi-Model Gen AI Certificate. The AI Infra and Operations Certificate can be valuable if you are interested in the DevOps side of AI. The Multimodal Gen AI certification covers systems that can work with multiple types of data such as text, images, and audio.
Whether you are just starting out your AI journey or trying to specialize in a particular area, these courses and certifications will work as a great resource for you. I will leave a link to all of them down below. If you want to know the fastest way to learn programming using AI, watch this video.
My name is Sahil and I will see you in the next one.