Generative AI Roadmap 2024 | Learn Gen AI in 6 Months | Build Your LLM, ChatGPT, Gemini, DALL-E, __?

hey guys welcome to analytics Vidya AI has become the silent architect of our rapidly evolving technology world today and if you think about it as more and more businesses are adopting or plan to adopt AI a career in this field will become extremely lucrative in the years to come so if you are someone who aspires to build a career in the domain of AI or more specifically generative AI then this video is for you in this video we are going to give you a 6 Monon step-by-step road map to learn generative AI in 2024 for Simplicity we have divided this road map into four broad levels of proficiency namely user super user developer and researcher in this road map we shall assume that you have user level proficiency already which is knowing how to use popular generative AI tools like chat GPT Bard or mid Journey Etc so as for our road map you will first go up from a user to a super user level by gaining a deeper understanding of generative AI through prompt engineering once done the next level is that of a developer where the objective is to First build and then deploy your generative AI applications using out of the box model apis and then you also go a notch ahead to fine-tune large language models or stable diffusion models for domain specific tasks and finally we have a researcher level where you strive to push the boundary of the Gen domain itself now at this point let's explore each of these roles in detail We Begin our journey with the goal of making you a super user of generative AI for that you need to learn prompt engineering and its various techniques and tricks this will take about a month guys let me call this out again to begin your uh super user Journey you must know how to use popular generative AI tools so understand what these tools are are what they enable you to do and have a hands Zone prompting experience of these tools now let's understand what you need to learn as a super user all right the goal of being a super user is to explore generative AI tools to its full potential and use these tools way more effectively so as a first step you need to dive into the theory of prompt engineering and the components of an effective well structured prompt it involves considering factors such as length of prompt structure providing context and Specific Instructions to shape the response of the AI tool you are quering we have made a detailed uh video on this particular topic you may check out the link in the description uh at this point you might be wondering is there a place where you can find good examples of prompts for uh different use cases answer is yes there are a bunch of online AI uh prompt communities like these where you can get good prompt examples to up your prompting game faster try them and see how a fine-tune prompt elicit good outputs from the large language model and diffusion model next I would uh recommend you to also learn about various prompt engineering techniques uh generative AI is a tool and uh these particular techniques can teach you how to use this tool in the best way possible again we have made a bunch of videos on various prompt engineering techniques do not forget to check them out as well by the end of the first month you will be able to enhanced prompts as a super user you may also upload your best prompts on various uh prompting communities and get feedback and further fine-tune your prompt thereafter that's all about super user now let's see the next step in our learning road map as a super user you are essentially an end user or consumer of generative AI but if you want to develop your own AI tools then you need to go deeper which we are calling developer level one Learning Journey at at this level should take about 2 months uh so let's get started again there are some prerequisits to this developer level one it includes basic understanding of uh programming language preferably python so you must know python as it would be a key to interacting with generative AI model via their apis or application program interfaces now let's move on to the developer level one road map all right if you are not familiar with API start by understanding what they are and how they work apis are nothing but a set of defined rules that enable different applications to communicate with each other next up specific to generative model study all the various API parameters like temperature setting Max tokens roles like system user and assistant Etc now focus on how these parameters can be used to control the behavior and length of the model's response then start exploring the apis of popular generative models like open AI chat GPD Google's G and Di 3 or open source models like Falcon metas Lama or stable diffusion which is a text to image model once you are done the next step is to master llm tools and Frameworks like Lang chain and llama index to build your own QA systems and retrieval augmented generation or rag systems apart from this learn parameter efficient fine tuning also called PFT PFT methods enable efficient adoption of pre-trained uh language models to various Downstream applications without fine-tuning all the models parameters plus remember most apis have usage limits which restrict the number of requests you can make in a certain time period they may also have commercial implications to familiarize yourself with these you should also uh understand the security aspect of using apis and error handling if anything goes wrong so once you are comfortable with prompt engineering and apis you can finally start building your own AI tools you may use Frameworks like streamlit or gradio to uh get started on building your own app we have already done a handful of jni uh projects and deployed them with the help of gradio app you may check them out Additionally you can also identify a different problem and uh use your understanding of apis and prompt engineering to implement your own Solution by creating an AI tool so this was the last step in the level one developer Journey our next step is to become a level two developer so let's continue as a level one developer you have interacted with the foundation models built AI tools by consuming deer apis now if you want to learn how to fine-tune Foundation models on your domain specific task making them more effective for your unique application that's where you need to level up to developer level two we'll spend 2 months here as well uh at the end of which you will be able to develop custom gen generative AI tools from scratch so for this developer level two again there's a set of prerequisits at this level you need uh a deeper understanding of python along with that you need a good understanding of uh these topics under probability and statistics linear algebra and calculus next dive deeper into machine learning concepts by learning the principles of supervis unsupervised and reinforcement learning also build a concrete knowledge of deep learning architectures and Frameworks thereafter you must also have the knowledge of fine-tuning foundational models like bird good understanding of attention mechanism Auto encoders and Gans at developer level two we now take a deeper dive into llms or stable defion models depending upon your area of interest if you are inclined towards the NLP domain explore popular large language models like gbd4 Gemini Pro or the open source models like Lama 2 understand their architecture training process and text generation mechanism and if you are more interested in computer vision domain focus on learning about foundational models in the computer vision space like stable diffusion models and their different types then learn about stable diffusion model architectures and their training process all this while participate in online discussions and forums to learn from others and clarify if you may have any doubt you now have the subject knowledge that you need now move on to the more Hands-On part as a first step Define a specific problem uh or a task that you want to solve using fine tuning under uh large language models select a foundation model that is suitable for your task while gp4 or Gemini Pro are popular choices they are not freely available open-source alternatives like llama 2 or falcon can be a good starting point for you for computer vision understand how to fine tune uh diffusion models for various uh Downstream use cases then focus on how to fine-tune stable diffusion models on your custom data set next comes setting up the fine-tuning environment you may use free platforms like Google collab or kaggle for your fine-tuning now if your task requires more compute consider uh using paid Cloud uh services like Google Cloud AWS or seur which often have free credits for uh new users as the last step start the F training process monitor the training and adjust parameters as necessary you may use techniques like PFT at this point once fine tuning is complete evaluate the model performance of your uh model using appropriate metric iterate and improve based on the results now comes the final step use your fine tune model to build a custom AI tool let's say your tool could be a medical diagnosis assistant that helps the doctors in diagnosing diseases test your tool thoroughly and further refine it based on the feedback using techniques like these by the end of this level you will have a deeper understanding of large language models and stable diffusion models the one that you opted for hands-on experience in fine-tuning these models and the ability to build custom AI tools on top of the fine-tuned model at this step you may also apply for various data science roles requiring expertise in Genna however if you still want to learn further the next stage of uh this Learning Journey is that of a researcher let's explore what that is the final stage of our generative AI learning road map begins at month 6 and uh there is no fixed limit to when it ends this is however an optional step those who aspire to contribute uh back to the field of generative AI can continue as researchers as a researcher you will uh delve into the intricacies of building generative models from scratch depending on the learning track you have chosen as a developer either NLP or comp computer vision let's break down each path if you had chosen the NLP track here is your road map learn and Implement attention models including key query value attention layer normalization and positional encoding gain handson experience in building your own GPT architecture from scratch dive deeper into reinforcement learning algorithms from basic to Advanced levels learn about proximal policy optimization or pop and Implement rlf from scratch and keep up with the latest trends and research that is happening in the generative AI for NLP participate in relevant online communities read research papers and attend conferences if you have chosen the computer vision track here is your road map learn and Implement uh diffusion models from scratch like stable diffusion these models are at the Forefront of generative AI for computer vision and building them from scratch will require a much deeper understanding of deep learning and computer vision and uh of course training Genera models for uh computer vision can be computationally very expensive so you might need to use high performance gpus or cloud-based uh services like Google Cloud awss for your continuous learning keep up with the latest trends and research in generative AI for computer vision participate in relevant online communities read research papers and attend conferences again as a researcher you should be able to contribute to the field of generative AI by building your own models and staying up to date with the latest research and if you are looking for a comprehensive path to become a generative AI expert even without leaving your current job you can enroll in our generative AI Pinnacle program as part of this you get a personalized learning road map cated just for you along with 200 plus hours of immersive learning experience 10 plus handson real world projects weekly one-on-one mentorship with Genera experts and you get a chance to master 26 plus generative AI tools and libraries link to the program is in the description part below you may check that out Additionally you may also join our analytics witha Community platform where you get uh data science and generative AI uh Community groups tailored to your interest opportunities to learn alongside your peers and above all you get a free access to live webinars and AMA sessions from industry experts so this is the end of the six-month long generative AI road map video uh whether you start as a super user developer or researcher there's a world of opportunities waiting for you out there and uh if you like this video subscribe to our analytics witha channel for more informative generative AI videos that are coming your way and uh please do not hesitate to share your queries or uh suggestions in the comment section below and we'll get back to you with our responses that's all we had for you today see you in the next video bye

Channel: Analytics Vidhya

Views: 10,060

Keywords: generative ai roadmap, genai projects, genai course, ai roadmap, ai roadmap for beginners, generative ai tutorial, gen ai projects, gen ai tutorial, genai use cases, ai jobs, genai jobs, how to learn ai, how to learn generative ai, chatgpt tutorial, chatgpt course, free ai course

Id: 4mYSR9m5NxQ

Length: 14min 8sec (848 seconds)

Published: Fri Dec 22 2023