Top-Down approach to get started with Generative AI and ML — Learning journey

AI is progressing at big steps, should we start learning theory or get into practice ASAP? That’s the question!

Photo by Nubia Navarro (nubikini) from Pexels

As software engineers or developers, we mainly focus on programming backend or web/mobile applications. Probably in the last two years you have felt overwhelmed with how fast Artificial Intelligence (AI) is making progress, and maybe you didn’t think you were going to see this AI revolution. Likely all the news around AI and Machine Learning (ML) piqued your curiosity, the ability for computers to learn on their own seemed like magic, but above all how fast those models are improving! However, you don’t know where to start or simply diving headfirst into algorithms and complex math is overwhelming for you. If this sounds familiar to you, keep reading since here we will discover the power of the Top-Down learning approach to begin our AI journey without failing in the attempt.

Artificial Intelligence can feel like a labyrinth — exciting possibilities branch out in every direction, but where do you begin? The answer might lie in your learning approach. Here, we explore the Top-Down approach with a pragmatic focus. Although this is just the beginning of my journey, I want to share experiences and leave some memories through the process, hoping they can help you to find your own direction in the world of AI.

As stated before, possibilities branch out in every direction. That’s why, before delving deeper into the Top-Down approach it’s important to do a quick research and define a goal. Let’s use my own example, and say my goal is to learn Generative AI because we can use all the tools and models available to leverage the power of AI. Machine Learning would be a step forward, to create or fine-tune models.

The Top-Down Approach: Big Picture First

While most of the books, posts and online courses use the Bottom-Top approach and encourage you to start learning linear algebra, calculus and statistics (we know this is pretty important). In the Top-Down approach you may focus on learning theoretical and general concepts like AI and its subsets (Machine Learning, Deep Learning, etc), prompt design and prompt engineering. And then you care about learning linear algebra, statistics, etc. This approach will be beneficial for you if you don’t have enough time or simply you want to start playing around AI as soon as possible. However, it’s ok if you want to try the Bottom-Up.

Thanks to the democratization of AI, nowadays it’s not reserved only for researchers and doctors in the area anymore, and anyone with an internet connection can have access to a tool that allows them to leverage the power of AI.

The Roadmap

Photo by Jaromír Kavan on Unsplash

Now, let me share my point of view about how a possible roadmap based on the Top-Down approach to get started with Generative AI will look like. However, it’s worth noting that this may change depending on your current knowledge and your goal:

1. Theoretical understanding of key concepts. e.g: AI, ML, DL, NLP, FM, LLM. The main idea here is to know those words or concepts that are used in almost every reading.
2. Prompt design/engineering including Zero, one, and few shots prompting, prompt chaining, etc.
3. Python fundamentals.
4. Use Generative AI APIs.
5. Jupyter Notebook.
6. If you already know Python you may want to go further and learn the basics of Pandas, Numpy and Matplotlib. This libraries are widely used in ML.
7. Understanding of key technical components in Generative AI. e.g: Vector Database, Fine-tuning, Embeddings, etc.
8. LangChain.
9. TensorFlow and Keras.

Only after the previous topics I would consider delving deeper into the ones that the Bottom-Top approach would have initially focused on. From my point of view the mathematics topics would be the most boring part. However, after understanding the relevance of mathematics in this field from a pragmatic standpoint; it may be easier and more interesting to learn. Therefore, some things might make more sense.

A long the journey the key is to balance the theory and the practice so that we don’t get bored and fall asleep each time we try to study (this happen to me when I start studying at 10pm after putting my daughter to bed 😬), we need to get hooked and embrace this process to increase the chances of success.

Takeaways

• Make a quick research to identify the state-of-the-art and the topics you need to learn. Remember that AI is moving really fast so don’t let too much water go under the bridge.
• Before you begin your learning journey, define a clear, measurable and achievable goal.
• Try to mix theory and practice at each stage so as not to become overwhelmed with so much theory. Furthermore, we can gain hands-on experience about what we‘re studying.

References

1. Juan Guillermo Gomez, Henry Ruiz, Snippets Tech Podcast: Machine Learning — Que es? — Con Henry Ruiz (2024), Spotify
2. Introduction to Generative AI Learning Path (2024), Google Cloud Skills Boost
3. Elvis Saravia, Prompt Engineering Guide (2024), DAIR.AI
4. Jason Brownlee, PhD (2020), 4-Steps to Get Started in Applied Machine Learning, Machine Learning Mastery
5. Adel Nehme, VP of Media at DataCamp (2023), How to Learn AI From Scratch in 2024, DataCamp Blog
6. What is Artificial Intelligence (AI)? (2024), Google Cloud Learn
7. Kaggle, Gain the skills you need to do independent data science projects (2024), Kaggle Learn

This is just the first of many contents that I plan to share about the Gen AI learning journey, so stay tuned and let me know in the comments which approach you would use and what would you change to the proposed roadmap. Thanks for reading!