Generative AI is everywhere and is evolving quicker than any other technology. If you want to learn GenAI in 2025, continue the reading :)
I started learning and working with Generative AI in Spring 2023, just after the launch of GPT-4 and the corresponding social boom it generated. At that time, I thought it was too late for me - experts would be everywhere, and there would be no room for me. However, I was clearly wrong.
During this year and a half, I have created my own recipe that has helped me to learn the topic from scratch and keep up with all the news and trends on the field.
1. Learn some Python 🐍
You don't need to be a Python expert to start your learning journey, but it will certainly allow you to quickly test what you are learning. There are many free courses on different formats you might find interesting. In my case, I went through this free course from exercism.org. Furthermore, try what you are learning about Python on your own. Jupyter Notebook is the perfect tool to experiment with your new acquired Python skills.
2. Understand how LLMs work 🔀
Large Language Models are at the core of any GenAI-based application. It is crucial to get the basics of how they are designed, created, trained, their underneath architecture, etc. If you understand how they work, you might understand their limitations. I think This one-hour introductory video from Andrej Karpathy is a masterpiece on the matter.
3. Prompt engineering 📝
Prompting is the based of building GenAI applications. It is the way you instruct the LLM to generate the content. Prompting is a very wide field, where you might learn new things every day. There are many free courses and websites that gather all about prompting, from basic topics to advanced ones like dynamic shots or prompt chaining. For example, webs like Prompt Engineering Guide or this one from Microsoft are quite interesting. Additionally, I found this 1-hour podcast about AI prompt engineering from Antrophic very stimulating: https://www.youtube.com/watch?v=T9aRN5JkmL8
4. Advanced topics ⬆️
Now that you know the basics it is time to explore more advanced topics like embeddings, RAG or function calling. These topics and techniques will open a bunch of new opportunities for your GenAI applications.
5. Use libraries 📚
Don't reinvent the wheel. It is important to understand the underlying concepts, but sometimes it is easier to just use a library that gives you a lot for free. Try those libraries which, in most cases, will simplify your journey in building GenAI applications. LangChain is the most famous one, but I think it is worth trying others, like Semantic Kernel, specially if you work in Microsoft ecosystem (Azure, .NET, etc).
6. Quality ✅
Quality is crucial in any application, and it is even more important on GenAI based application, to keep content created relevant and reduce hallucinations. Understand how you can create evaluation systems for your application. Also, concepts like LLM as a judge might be very interesting in some scenarios.
7. Stay updated 💡
The rapid pace of changes and improvements in GenAI is stunning. You might feel lost just after couple of weeks of disconnection. Stay updated with the latest trends and news through newsletters. I follow few newsletters that cover commercial and non-technical news, such as Artificial Ignorance and Why Try AI. I also find the Ahead of AI blog useful, as it delves into more technical and deep concepts. Additionally, the AI Explained YouTube channel is quite entertaining and informative.
8. Experiment 🧪
Work on your own ideas, build prototypes, and don't hesitate to discard them and try new features. Theory is important, but experience is crucial. Hands-on experiments is by far the best way to learn new concepts.
These 8 steps are a perfect summary of what has worked for me. I hope you have found this reading useful and interesting. Let me know in the comments how you have learned GenAI or any content, blog or video you have found useful on the topic.
You can also follow me on LinkedIn where I publish GenAI content as well.
Top comments (0)