Hello Everyone,
There’s so much to learn and I’m a big believer that nobody is an expert in Generative AI although obviously the people most recently educated in ML have a distinct advantage to be educators of the tech aspects of Generative AI.
Whatever your level or industry and field of interest around Gen AI, there are people who you can follow. Currently I’m working on a list of who to follow that I will publish soon.
With the Generative AI boom, there is increasingly a lot of great educational material about AI online. This includes a lot of speciality content around Generative AI specifically and emerging educational AI creators. I’m a fan of many of them, and one of my favorites is Aishwarya Naresh Reganti. I’ve rarely come across someone who offers so much value for free.
She has tons of material available on Github and on her posts. The guest contributor of today, did a Masters at Carnegie Mellon University, known for its strong CS and AI programs. In fact, Carnegie Mellon University was ranked joint #1 in computer science in 2022-2023 in the world.
📶 From our sponsor: 📶
🎬 AI Video Repurposing — Extract the most engaging and impactful clips from your long-form videos using Munch AI.
Get 50% OFF all Munch plans and grow your business with dozens of clips!
Munch offers groundbreaking solutions for businesses looking to stand out on social media, leveraging AI to transform long-form videos into dozens of viral social clips.
For AI Supremacy readers, get 50% off your first month on all Munch plans. Code: AIREP50
I really admire educators in AI of all types and I have a lot of readers from India and Asia generally. South Asia and India specifically is such a huge immigration talent driver for machine learning engineers, scientists and researchers in Silicon Valley and the U.S. broadly speaking. It’s incredible to behold how generous some of these younger engineers and ML scientists are as well.
Aisha is now Gen AI tech lead at AWS and a visiting lecturer at MIT. In some ways this means she’s at the heart of the Generative AI movement. From Bangalore to San Jose, she’s a Top Voice in Generative AI on LinkedIn.
Check out her Generative AI learning resources on the Github she manages here:
https://github.com/aishwaryanr/awesome-generative-ai-guide
She provides so much for people learning about Generative AI and ML on her LinkedIn posts.
I asked her for a basic walkthrough to Generative AI, and this is her article. She’s a Generative AI tech lead at AWS, Amazon.
“How do LLMs work? Here’s a beginner-friendly video in under 90 seconds! She’s so full of resources and helpful guides!”
🎓 “Generative AI Genius” is a 20-day introductory course with short videos/reels designed to help you break into generative AI. Post | Generative AI Genius Course | Register (Starts July 8th, 2024).
Introduction to Multimodal LLMs: Post | Gentle Introduction
I’ll be doing more educational and AI tool deep dives for my premium subscribers so show your support and don’t miss out.
Educational Resources by the Author
5-Day LLM Foundations Roadmap 2024 (link)
List of Free GenAI Courses (link)
Applied LLMs Mastery 2024 Course Content (link)
3-Day RAG roadmap (link)
“Retrieval Augmented Generation (RAG) is a super cool and important concept for LLMs that you should know!”
Getting Started in GenAI: A Beginner’s Guide
By Aishwarya Naresh Reganti, May, 2024.
The rising tide of interest in Generative AI is undeniable. Its potential to streamline numerous tasks has captured the imagination of many. For the average person with even a modest familiarity with computers, Generative AI holds the promise of significantly reducing the effort required for various tasks.
However, navigating the world of Generative AI can seem like a maze of hype and complexity. Where do you even start? How do you ensure that exploring this field will bring you real benefits? The sheer amount of information out there can leave you feeling overwhelmed and disconnected.
When folks ask me where to begin, I often advise them to take a step back before diving into Generative AI. It’s crucial to pause and reflect on why you’re interested in it. There’s a vast sea of knowledge, but you only need to focus on what will help you achieve your goals. In this blog, we streamline the process by initially pinpointing your goals and determining your audience type using the categories outlined below. You’ll definitely find yourself fitting into one of these three categories.
The infographic above delineates the various audience types and the corresponding skills necessary:
Non-technical Individuals: You may not be a tech wizard, but you’re comfortable with computers and eager to explore how Generative AI can enhance your daily tasks.
Tech business leaders and practitioners: You’re no stranger to technology, but Generative AI remains a bit of a mystery. You want to harness Generative AI to develop real-world products and understand the challenges in doing so.
AI/ML Specialists: You’re well-versed in machine learning, but Generative AI is uncharted territory for you. You’re keen to dive into research, build models, and refine them.
Once you’ve identified the group that resonates with you, you can direct your attention towards gaining the skills and knowledge that align with your interests. For example, if you’re a business leader, your focus may revolve around understanding deployment logistics such as selecting AI vendors and optimizing metrics like latency and cost. Conversely, if you’re an AI specialist, your emphasis might lean more towards crafting advanced Generative AI models and similar endeavors.
So, before you go further into this blog, take a moment to determine which category resonates with you best. It’ll serve as your compass as you navigate the super interesting world of Generative AI!
Authors Top LinkedIn Posts:
What are the next big trends in LLM research?
My repository is now updated with over 85 free Gen AI courses
Building LLM applications becomes simple when
Create your own LLM RAG application in just 3 days
Non-technical Individuals
If you’re eager to make the most out of generative AI technology, there are two key concepts you should grasp:
Prompt Engineering: This involves understanding how to effectively interact with generative AI models or LLMs (Large Language Models) to obtain the best responses from them. You’ve probably engaged with technologies like ChatGPT or Claude, which may seem like smart conversational bots. However, the way you communicate with them can significantly impact their responses. For example, did you know that asking LLMs to explain their reasoning process step by step can lead to more accurate results, particularly in complex problem-solving scenarios? If you frequently interact with LLMs and aim to optimize their responses, learning about prompt engineering is essential. You can access some free courses on prompt engineering in my repository; feel free to choose any and get started.
Application Usage: Another crucial aspect of generative AI is its application beyond mere conversation. You can leverage generative AI to develop various applications, not limited to chat interfaces. For instance, you can use it to automatically generate meeting notes by providing access to call audio, create travel itineraries, or even transform selfies into professional headshots. The possibilities are vast and varied. Here are a few resources showcasing existing generative AI applications. Take a look at them to explore the types of applications that could be beneficial for you:
https://github.com/ai-collection/ai-collection
https://github.com/steven2358/awesome-generative-ai
Tech business leaders and practitioners
If you’re a business leader or practitioner in the tech industry, you’ve likely heard about the significant impact of generative AI tools and are keen to integrate them to enhance your current workflows. Here are some key areas that should be your focus:
Deployment: Deploying generative AI systems involves adhering to standard ML pipeline practices, but with unique challenges and considerations in mind. The LLM technology stack, as illustrated in the diagram below by Sequoia Capital, illustrates various components essential for constructing generative AI pipelines. This includes LLM providers, vector databases for RAG applications, and frameworks that offer abstractions facilitating application development. Collectively, these components contribute to the successful deployment of generative AI applications. They facilitate efficient LLM interaction and the delivery of impactful outcomes to end-users. Therefore, it’s essential to focus on comprehending the stack required for your application by diving deep into integrations and understanding how they collaborate with each other.
Image Source: https://www.sequoiacap.com/article/llm-stack-perspective/
2. Market Landscape: Whether you’re a business leader or a practitioner, understanding what the market offers and how to leverage it is essential. For leaders, this knowledge provides a competitive edge, while practitioners benefit from knowing the optimal tools to use. A useful approach is to identify the parameters you’re optimizing for and explore options along those dimensions. For example, if cost is a concern, consider open-source or low-cost LLM tools or LLMs themselves. If size is critical for on-device applications, research the best-performing small models available. Similarly, if context length is a priority, seek out model providers with the longest length. The landscape will vary based on your optimization goals, but comprehending it is key.
3. Evaluation: A critical element in developing generative AI products is establishing a robust evaluation pipeline. This pipeline is instrumental in assessing both task-specific metrics and the business objectives you’ve set. Evaluation metrics for your task can be of different dimensions
Evaluation of the pipeline: Assessing inputs, outputs, latency, costs, and aspects that relate to the pipeline and not the generative AI model itself
Model Evaluation: Evaluating task metrics, benchmarks, human alignment etc. of the model.
Here are some resources to get started: Free courses on evaluation in my repository, A beginner article by Microsoft.
4. Challenges: Understanding the unique challenges that come with LLMs is crucial before deploying them. These challenges include common issues such as hallucinations, where the model generates content that doesn’t align with reality, and adversarial attacks, which can cause LLMs to produce responses that deviate from expected human behavior or even limitations such as the model’s restricted context length or its performance degradation with longer contexts pose significant challenges. Some common LLM challenges are presented in this paper, as depicted in the image below. It’s important to grasp these challenges thoroughly and develop effective strategies to address them. Always stay vigilant about potential issues to ensure smooth deployment and operation of LLMs.
Image Source: Challenges and Applications of Large Language Models (pdf)
AI/ML Specialists
If you want to dive deep into LLM research and development, it’s important to recognize that it’s not a simple task. It requires years of ML knowledge to grasp the intricacies of building LLMs. Before diving in, it’s crucial to have a solid understanding of the foundational concepts. But let’s say you’re already familiar with how ML models operate and want to explore generative models further. Here are some key areas to focus on:
Training Paradigms: The latest LLMs may share similar architectures with older models like BERT, but what sets them apart is their training paradigms and data. This includes factors such as the type and size of data used, subtle architectural modifications, and novel training techniques like RLHF instruction tuning etc. Understanding these details is essential if you intend to develop your own model. Exploring research papers on popular open-source models like Llama, Gemma, and Phi can provide valuable insights into their training methodologies.
Latest Architectures: Building on the previous point, it’s crucial to stay updated on the latest advancements in model architectures. For example, recent innovations such as MoE (Mixture of Experts) architectures have demonstrated significant performance improvements. Similarly, architectures like RWKV and Mamba state-space models are pushing the boundaries of what’s possible. Understanding these architectures and their implications can give you a clear idea of upcoming trends and how you can contribute to them.
Emerging Research: In addition to understanding foundational concepts and staying abreast of the latest architectures, it’s essential to keep an eye on emerging research trends. This involves staying informed about cutting-edge developments and innovations in the field. For example, trends like small language models, which offer specific task-solving capabilities with low compute overhead, or multimodal models, which can process and generate different modalities, are gaining traction. Reading research papers and keeping tabs on emerging trends can help you stay at the forefront of the field and make informed decisions while building models.
Resources
Here are a few resources to help you stay updated on the latest research:
HuggingFace Daily Paper list: https://huggingface.co/papers
DAIR.AI papers of the week: https://github.com/dair-ai/ML-Papers-of-the-Week
Biography
Aishwarya works as a tech lead at the AWS-Generative AI Innovation Center in California, where she leads projects aimed at building production-ready generative AI applications for medium to large-sized businesses. With over 8 years of experience in machine learning, Aishwarya has published 30+ research papers in top AI conferences and mentored numerous graduate students. She actively collaborates with research labs and professors from institutions like Stanford University, University of Michigan, and University of South Carolina on projects related to LLMs, graph models and generative AI.
Outside her professional and academic pursuits, Aishwarya actively contributes to education through various channels. She offers several courses online, with over 3000 individuals having taken them already, and serves as a visiting lecturer at institutions like University of Oxford and Massachusetts Institute of Technology (MIT). Additionally, she co-founded The LevelUp Org in 2022, a tech mentoring community dedicated to assisting newcomers in the field through mentorship programs and career-oriented events. A recognized industry expert and thought leader, Aishwarya frequently speaks at various industry conferences like ODSC, WomenTech Network, ReWork, and AI4, and has presented research at top-tier AI research conferences including EMNLP, AAAI, and CVPR.
Editor’s Resources to Learn from Scratch
Generative AI Guide on Github
Generative AI for Beginners, Microsoft on Github
Coursera: AI For Everyone (by DeepLearning.ai)
Understanding Large Language Models
Intro to Large Language Models
Andrej releases very helpful videos periodically on his YouTube.
What Is an AI Anyway?
A TedTalk by Mustafa Suleyman
Is AI turning into something totally new? Mustafa also had a relatively new book out that many analysts and people in the space recommend.
🙋♂️ Your Turn
What Podcasts, YouTube creators or Newsletters around Generative AI do you recommend?
Read More in AI Supremacy
