Generative AI: Getting Started

1. Generative AI - Course Introduction: In this video course, learn about generative AI and its concepts. Generative AI is a subset of deep learning. It uses AI neural networks and can process both labeled and unlabeled data using supervised, unsupervised and semi suupervised methods. It refers to a class of artificial intelligence models and algorithms designed to create new content. These models can generate text, images, music, and other forms of data. That mimic human created content. Generative AI applications are built on top of large language models. These large language models are deep learning models. With Generative AVA chat boards, such as CA GPT, Google Gemini, Microsoft Co Pilate and others. You can easily create images like logos, banners, et cetera. Scan images and search PDF documents. Also write professional e mails, blogs, and articles in seconds. These chat boards can also teach you coding. Write advertisements for you. Fix the grammar, plan your vocation, and B your everyday AI assistant. The following lessons are covered in this course, Let's start with the first lesson. 2. AI vs ML vs DS vs DL: In this lesson, we will understand the difference between artificial intelligence, data science, machine learning, and deep learning. We will also see how these are related to each other. With that, we will also understand that how generative VA is related to these terms. Let's start. As I told you that I'll be discussing this first. Why? Because our generative A is also part of this AI. AI is a superset, as you can see in this wind diagram. It's a superset. It includes your machine learning, deep learning, and data science. But the ecosystem of data science also exceeds AA. What is AI A means creating smart machines to mimic human behavior? Or we can say it refers to the simulation of human intelligence in machines that are basically programmed to think and learn like humans. You must have seen AA in a lot of domains these days because you can easily analyze large amounts of data, recognize patterns, and make decisions. It is mostly used in healthcare finance, transportation, and entertainment fields. These days. Then comes your machine learning, which is a subset of artificial intelligence that is AA. Machine learning is a subset of AA, as I told before. And it is used to build a model based on training data to make predictions. Using machine learning, you can build a model to Make predictions, let's say, to predict the winner of this world cup. It focuses on developing algorithms and statistic models that enables a computer to learn from and make predictions or decisions based on data without being explicitly programmed to do so. Its techniques includes your supervised, unsupervised, semi supervised and reinforcement learning. It is also used in various fields such as image and speech recognition NLP, that is natural language processing. Forecasting medical diagnosis and others. Now comes your data science. Data science is the subset of AA, as I told above. It is an area of statistic, scientific methods, et c to extract meaning and insights from data. So I'll give an example. Let's say you went to Instagram and you liked some car videos like MG, Kia, Honda, Tesla. What will happen? You gave your data to Instagram that I like such videos. Such Instagram reels, Instagram channels, Instagram accounts. So what will happen? The next time you open Instagram, The Instagram will automatically pitch you with such reels, such post, let's say some discounts on cars. So how these things happened? All these things happened due to data signs because It extract meaning and insights from the data. Now, let's say a car company wants to approach some people who love cars. Whenever they'll add a sponsored post or story on Instagram, they know that These number of people like car videos, so the same thing will be pitched to them. What Data science did, they connected the client with the company. In this way, both parties got benefited, the client got that discount, and the company sold their product. So that's the value of data science. We say, data is the new le because an unprocessed data is of no use. Similarly an OL is of no use if it is not processed properly. Therefore, data is processed and meaningful insights are generated. Now comes your deep learning, deep learning, you can consider as a subset of machine learning. According to the n diagram, you can see. It is a class of machine learning algorithms to solve complex problems. It focuses on using artificial neural networks with multiple layers to model and understand complex patterns in data. Deep learning algorithms are inspired by the structure and function of the human brain, specifically, it's interconnected network of neurons. Why we are discussing this? Because generative VA is a part of deep learning. 3. Deep Learning Types: In this lesson, we will understand the types of deep learning. This will also help us in understanding that how generative a is related to deep learning. We will also see an example. Let's see. Dep learning types include your discriminative as well as generative. Previously, we all discussed about this, let's say to classify between a dog or a cat from a bunch of images from some images. Okay. Discriminative deep learning is used to classify or predict. It discriminates between different kinds of data instances. Let's say you have some images and you want to classify them as a dog or a cat, so it will be able to discriminate between them, and we'll predict that which of them is the pick of a dog or a cat. But generative AI is a completely different concept. It will generate new data that is similar to data it was trained on. It generates new data instances. That means in this case, it will generate a new cat Mage. Let's say you will upload your pick and it will generate your AI of TA. Or let's say you added a text prompt. Let's say you want to know about anything related to cricket. So you'll ask the prompt, and it will generate new data or content that resembles the original data it was trained on. Okay? 4. What is Generative AI: In this lesson, we will understand what is generative VA? We will also understand its process that how it generates new content. Let's see. Now, since we discussed about generative I, I told you that it is a part of deep learning. You can see. Generative VA is a subset of deep learning. It uses AA neural networks and can process both labeled and unlabeled data. That means as before the types of machine learning, a supervised unsupervised and semi supervised methods. GI, that means generative AI is a class of AI models. That is designed to create new content. It can generate not only text, but images, music, and other forms of data. It is built on large language models. We will also discuss large language models later. These LLMs, that means large language models are deep learning models. This is the process of generative AI. I told you it creates new content based on what it learned from existing content. That means the data it was trained on. Here, training means learning from existing content. It will create a statistical model. That will be used to predict an expected response. When you type a prompt. When a prompt is typed, this generative I will use the statistical model. To generate new content in the form of text, images, music, video, task, and others. 5. Techniques for implementing Generative AI: In this lesson, we will understand some techniques for implementing generative AA. You can also consider it as the approaches or the generative AI models. You must have heard about GPT three and GP four models of pen AA. These are also based on these techniques. Let's see. Now, let's see the techniques for implementing generative A, or you can also consider it as generative AI models. The first one is GAs. Generative adversarial networks. Under this, two neural networks are trained simultaneously. The first one is a generator network, and the second one is a discriminator network. The generator creates data while the discriminator evaluates it. You can say the generator network lens to generate data samples such as images or text that resemble your training data. While the discriminator network learns to distinguish between real data samples and those generated by the generator. The second one is variational auto encoders. These are basically used for encoding and reconstructing data. It is also a type of generative model used in machine learning and deep learning. The variational auto encoders, can generate new data that's similar to the input data they have been trained on. You can use it to create new images that resemble a given data set. VAs are used in generative modeling, data compression, et cetera. Now, let us see transformer based models. Using these, we can easily handle large sequences of data, particularly in NLP task. This is the topic we'll be discussing. Because this is behind some of the most advanced language models like AI GPD three and GPD four. Two of the most powerful generative a models. These are based on the transformer architecture. The transformer architecture was coined by Indian in 2017. These models are used to generate human like text. It can also help you with coding tasks and translate from one language to another. Let's learn more about this. The GPT means generative pre trained transformer. That's why we are discussing this topic. 6. Generative AI - Transformers: In this lesson, we will understand what is a transformer. We will see what is a transformer model, its architecture, who coined it with that we will also see its process. While using transformers, you may run into an issue called halucination. We will also cover what are hallucinations and why it can happen. Let's start with the transformers concept. Here comes your transformers. It is a type of generative a model. That is a type of generative a model called transformer model. You can consider the power of generative A comes from the use of these transformers. I told you it was coined by Indian in 2017, Ashish asi. Okay. It helped in actually laying the foundation for advancements in the field of NLP and machine learning. Okay. The transformers include encoder and decoder. I'll also give an example later. The encoder will encode the input sequence. Let's say you have a text in Spanish language, and you want to convert it in English language. What will happen? The encoder will encode the input sequence and pass it to the decoder, which will learn how to decode the representations for relevant task. Let's see. Here is the process I told you encoder decoder. It is the main component of the transformer architecture. L et's say we have a text my name is amet and Spanish language. What will happen with the transformers? It will get first encoded. That means the encoder will include your self attention and feed forward mechanisms. What will happen? Every word will be related to every other word in the input sequence. This will allow the process to focus on the key words in this. Now, the next mechanism feed forward, what will happen? This will further refine the understanding of each word, and it will be passed to the decoder. Further, the decoder will will generate the Spanish text in English language. That means a text in Spanish processed into its English equent using transformers. An issue may arise while using transformers. That means haucinations. You must have heard about AA showing irrelevant results, misleading results, grammatical issues. All these come under hallucinations. Here you can see misleading results. Alucinations are words or phrases that are generated by the model. That are often nonsensical or grammatically incorrect. It can be due to various factors. Let's say that data is noisy. It is not having enough context or the model is not trained on enough data. So Illustinations, since those are misleading results, make the output text difficult to understand. 7. Large Language Models (LLMs) and its use cases: In this lesson, we will learn about LLMs. That is large language models. Whenever you discuss about generative I, then this topic will always be considered. Both LLM and generative VAs are subsets of deep learning. Let us understand what are LLMs, and we will also discuss a type or you can consider a use case of LLMs. Let's see. Okay. Now we'll be discussing about large language models. I told you that generative VA is a part of deep learning, and LLMs are also a part of deep learning. Both are related. LLMs are also a subset of deep learning. As I just said, Okay, you must have heard about CAT GPT, copilot, Google Gemin, that means bad mid journey. LLMs are AI models. You can consider that power, all these chat bots. LLMs are large language models. That means large general purpose language models. That can be pre trained and then fine tuned for specific purposes. You can pretrain LLM with a large dataset, and fine tune means to fine tune it with a particular M with a smaller dataset from that large dataset. LLMs also represent a class of AI models that is used to understand and generate human like text, or you can say it provides an engine that powers the AHd bot. You AHd bots are based on These LLMs. These LLMs will allow your chat bot to easily create naturally phrased recommendations so that the content is generated by generative AI according to your personalized recommendation. That's why LLM is considered as the backbone of AHd bots, all the AHd bots. Now let us see a scenario or a use case. The large language models are trained with petabytes of data and generate billions of parameters. To solve different task. These tasks can be sentence completion, text classification, language translation. We can see this example of Palm PLM. It is a transformer based large language model. Google just announced Palm two also. It is a pathways language model, a 540 billion parameters, that is a larger training data set with a large number of parameters. It is also a transformer model. I just told you that transformer model includes your encoders and decoders. I discussed this before. So The speciality of LLMs are that it can still obtain a greater or decent performance with little domain training data. So it can be used for few shot or even zero shot scenarios. So these two scenarios, if you'll learn more about LLM and all these models, you will be getting such terms again and again. So let me explain it quickly. If you're training a model with less data with minimal amount of data, then it would be called few shot as the name suggests. And what about zero shot? It means a model can recognize things that have not been taught in the training before. That means zero shot, nothing. LLM the performance of LLM grows when you add more data and parameters. Here we just saw f 40 billion parameters. We can learn more about Palm later. It is considered as a next generation language model. With the enhanced multilingual reasoning and coding capabilities. Okay. Google also announced Audio Palm for speech to speech translation in June 2023. 8. Generative AI - Applications & Challenges: In this lesson, we will learn, what are the applications and challenges of generative VA? We can easily generate content images, logos, banners, as well as summarizing PDF using generative VA. But we should also understand the challenges behind it. Because this is also a topic to cover since generative VA is used for some unethical purposes also. Let's see. Now, let us see some applications and challenges of Generative AI. We all know that generative AA can be used to create content, proof re date, we can write e mails. We can also create characters, three D images, games. We can create complete landscapes and scenarios. It can also be used by artists and designers easily. You can also generate logos, banners, social media post. What are the challenges I would like to discuss this more? We saw that these generative AI models are basically considered to have ethical concerns, quality control, biasness. Also the images you are generating, the texts you're generating. Some people say that it can have copyright issues or even on YouTube, they are asking that is your video generated by AA or not? So you can explicitly mention it. Also on Instagram, there is an option to add your AI label. Okay. With that, one of the challenge or issue with GNI. Here it is. On Google Gini, once it was showing some misleading results, like, people can eat rocks and they can glue pizza. So someone surged and Google AI surg revealed the following results. So these are very cary. When you'll learn about Google Gemini. Now they are showing a disclaimer that if you're creating a fitness plan or a meal plan using these Had birds, there is a disclaimer that you need to contact a registered dietitian or a fitness expert before following our answers before following what the prompt result is. These things are really important. 9. Generative AI - Chatbots (Model Types): In this lesson, we will learn about the generative via chat bot model types. You must have heard about the text to text model, text to image model, text to video, text to music model. Let's see what are these. Now, let us see the model types, so this will also cover your EI chat bots currently. Okay. Text two text, we all know, OP EI Cat GPT, Microsoft Co Pilot, and Google GMI, we'll be typing a text prompt to generate an e mail to generate an article to generate a block. With that, we can use text two images on the Dali model and Md Jury. Dali model is now part of Microsoft Copilot, so you will be getting around 15 boosts in a day within the free version of copilot so that you can work with Daly. Text to video, Open A Sa, and we now also have Kling and Luma AI Dream machine introduced today. So you can try and generate text to video now. Easily. Text two songs, you must have heard about creating songs using text prompts using a two line prompt with So AA. Okay. You can easily achieve this with text two songs model. Then comes your text to task like software agents, virtual assistance, automation. So Microsoft came up with the co pilot PCs, the copilot studio also as in virtual assistant to easier work. 10. Generative AI - Features & Examples: In this lesson, we will learn about generative via chat board features, as well as some examples. With the features, we will see some examples related to text to image to image, as well as text to video. Let's start. Now the features of AHd bots, I have shown these chat bots before. Now the features, I have just amalgamated all of them. You can easily create logo banners. You can also use it to code, fix your code, generate syntax. With that, you can also upload and scan images. This means that if you're having an image and you want the AH adb to read it, to scan it. That What does this image include? So our tutorial also includes that complete use case. You can directly ask that what does this image include? By aplouding EPEC. Okay. With that, you can easily work search and scan PDF documents. So 20 pages, ten, 20, 50 pages documents PDF documents can be scanned within seconds within less than a minute. Definitely, it will save your time. It will save at least five to seven 8 hours of your work. If you will summarize a 30 to 40 page document. You can write e mail blogs and articles. You can also set the tone, the number of words you want. If you want storytelling for your article, you can easily add it. With that, you can easily find jobs, create resumes, cover letters from that resume. I told before that we can also use it for coding. You can also write advertisements. You can generate product timelines. Fix the grammar proof, read, your content completely. Plan your vacations, generate hotel recommendations completely. You can write a meal plan, a fitness plan based on your recommendation. Let's say if someone wants a fitness plan without using gym equipment, exercise plan, your chat board can also do it for you. With that, you can also get gift ideas. In fact, in Google Gemini, you'll also get images and the link to get some gift. Let's say you want a gift for a kid aged six. You can mention I want a gift for a boy kid aged six. Then you'll get relevant results. These are some examples. I generated them using text image model. Let's say Dali, Dali is a part of copilot now. You can easily generate images by just typing one to two line prompts. So this is how I generated it. I generated this three D line with sunglasses, a robo image, a dog playing on a road, and these also. This was text image model. You can also generate them using mid Journey, encraft, Daly, and others. This is Image two Mage model. So I took this image from the Internet. So these are the images of cricketers when they were kids, and this is my image generated by an image two image model. 11. What are Prompts: In this lesson, we will learn what are prompts. With that, we will also learn what is prompt engineering, as well as who are prompt engineers. The role of prompt engineers are becoming popular. Let's see. Now, let's see what is a prompt. Prompt is basically the input that a user types. I told you text two image text to text text to video. So those were the prompts. What will happen when a user will type the input, it will go to the AI model to get a specific response. A new response will be generated. That's the purpose of generative A to generate new content in the form of text images videos. Okay. You can also consider it as a query. It describes the tasks that an AA should perform. Let's say you t I want to write an e mail to my boss for five days leave. Okay, so the response will get generated. What is prompt engineering? It means crafting specific instructions that can be understood by the AA model. And to get responses in real time. That means what you will type and the result will get generated immediately. I've just shown you the images I generated. Also you can generate text from your text prompt. Now, what do prompt engineers do? So the role of prompt engineers are becoming popular because if you know how to craft prompts properly, you can easily generate results because a lot of these prompts are limited. Copilot provided daily, but you only get 15 boosts in a day. Also for text, you have some tokens, Token, you can consider half a word or 0.75 of a word. Okay. So those are also limited for a day. Those keep on changing. 12. Popular AI Chatbots: In this lesson, we will learn about some popular AI chat bots. Some of them are widely used. The first chat bot was introduced by OPA. That is Cgb. Then came your copilot and Google Gemini. Guys, the following are some popular AI chat bots. The first one was introduced by OPA, that is Open Gb. O pene also introduced Daly for images, and they also introduced Open SRA for videos. Microsoft launched copilot. A lot of people don't know that CAD GPT supported by Microsoft. It got funded by Microsoft. That's why in Copilo, now we have Open Dal for image generation. Okay, Google Gemini was known as bad and it was obviously developed by Google. So let's see the layout quickly. These are the links wherein you can access them. Here are the layouts. The following. The first one is for Chart GPT. Okay. This is the free version. If you'll type any prompt, let's say, so you can now see you have GPT four for free. Here it is, GPT four f free with limited prompts and image generation. Then we have our copilot. These are copilot GPT. Let's say you want images, you can click here, and you can generate logos images. The website is copilot microsoft.com. The last one is Google Gemini. Okay? These are the suggested prompts, and these are the prompts I wrote. Under settings, you can select the dark theme or you can also select extensions to work on tuks. Okay guys, Guys, we saw how we can easily work around Generative VA, what is generative VA, its models. We also learned about its features and the types, the transformer model, its process, the process of generative VA, and we also worked around some great examples to understand the text text to image, as well as text to video models. Thank you for watching the video. 13. ChatGPT-4o Quick Overview and Use Cases (Prompts): So the g4o is here. O stands for Omni. It includes your access to audio vision and text in real time. Here it is O for Omni, and it accepts as input any combination of text audio and image. You can scan images as well as improvement on text in non English languages. Also, the APA is 50% cheaper. Let's see how to access it. Just go to chart. I already locked in for the 3.5 version, so it is directly giving me an option to try it now. I told you it can understand images, can browse the web and speaks more languages. I'll click Try it now. Here you can see g4o. Now, let us start whether it is connected to the Internet or not. So GPT 3.5 wasn't connected with the Internet. But to this claims, that an internet connection is there for G four. What is the temperature today in? Deli India? Here it is the current temperature is visible. That means it is connected to the Internet. Okay. From here, you can change the model also. Right now, we are in GBT four. Now let me generate a logo. Create a logo for an for an online shopping company with the text one stop shopping destination. Let's see whether it will generate a logo or not. No, it is not providing. No, let us upload an image. Here it is type. What is this image about? Let's see, will it be able to check the image, scan the image or not? Okay, it's a Laptop smart phone apple. Fine. We can also learn about this. Now let us upload another image. I'll upload my image open? Let's see. Submit. Okay. It has represented it. Now, I'll click on this shoe image. I'll type. Have you seen this before? So I'm just scanning it. Submit. It was able to guess it perfectly the name of the shoe. Okay. No solve this linear equation. I'll direct solve. It is showing the steps also. Okay. Here is the answer. You can see the steps. Now you can see when I'll click here, you have reached your phyplod limit. You can upgrade to chat GPT plus or try again. Okay. Let's write an article on a current topic. Write an article on IPL, Indian Premier League is here, and we are nearing its end. Ten to 15 matches are remaining. Let's say how much it is updated. It is searching the news. Okay. Here it is. Let me know about the current matches of IPL 2024. Of, it should be off. I edited it. Similarly, you can also edit. I saved it and edited. You have reached our limit of messages. Please try again. Region. This is how I demonstrated the 4.0 version, the following. You can rename it. G four First impressions on Laptop. Okay. So definitely, if you want the voice access also, you need to go to your mobile phones and download Chat GPD there. Download and install Chat GPD there for the GPT 40 version. So guys, we saw some first impressions of GPT four. Thank you for watching.