So you want to do AI

How to learn about AI in a world that's supersaturated with hype

Mar 28, 2025

I used to say "No LLMs, No AI" when pursuing work or projects. It was too early. It was changing too much. Too much hype. Now, if you're saying that, you're absolutely getting left behind.

As an industry, we see better where AI fits and where it doesn't. We see the opportunities to leverage it. And not only that, but if you're not leveraging it, you are getting left behind by your coworkers and others at your level that know how to use it.

Learning how to leverage AI is not an overnight task, it's taken me time, and continues to take me time. In this post, my goal is provide you a few avenues for how to learn about AI. I'm not an expert, but I am someone that touches it daily and exposes myself to various ways of learning its capabilities.

I will cover a few paths to enlightenment:

Reading
Toying
Practicing
Talking

Reading

Remember that Goodreads reading list you have sitting and waiting for you? It might be time to reorder some of those books and adjust your priority.

Reading long-form material (books) is absolutely the best way, in my experience, to expose yourself to new topics. The field is moving fast, but books will introduce you to concepts that are the foundation of AI. They also have the space and breath to be able to explain to you the very concepts they introduce you to, should they chose to. And since books go through rounds of revisions, you're reading something that has been reviewed by people much smarter than either you or I.

Some books I can recommend at the moment:

AI Engineering, Chip Huyen (somewhat technical)
Why Machines Will Never Rule the World, Jobst Landgrebe and Barry Smith (more technical at times)

Toying

My dad always told me when I was kid the best way to learn something is to break it and put it back together. That's not necessarily true in this instance but the sentiment matches; don't need to take it so seriously. Play around with it.

Some ideas:

Ask ChatGPT how it works
Use Gemini 2.0 Deep Research or Perplexity. Watch how the models reason through what you've asked. This will give you a better idea of what they're doing.
Replit. It'll build a whole app for you basically in a vibe coding session. Fun to watch.
Watching YouTube videos of AI-based projects falls into the toying category for me. You can relax and listen without having to be hands on keyboard.
Using Cursor or VSCode to help you create code while you're working on something
Use any of the models to help teach you a new concept, have it quiz you after, and help iterate through multiple quizzes proving you know the material (but be careful of hallucinations and the model actually giving you the wrong answer)
- I do this when I'm trying to learn new data architecture or coding pattern concepts. Don't give me the code. Let me understand what's underneath first. Then start giving me little pieces of code. AI is a great iterative teacher and tutor.

Practicing

This will be a bit more hands-on than toying. This means you're building something that's going to have AI in the DNA.

Experiment with things like Temperature, controlling how random or consistent the model results will be.

Some project ideas:

Youtube Video Summarization: user passes through a URL, have a service that downloads the video, transcribes, and summarizes the key points. *
Large text summarization: Same as above, just skipping the Youtube downloading piece. Use a model like gpt4, hit the OpenAI API, and pass the text to the prompt
Custom Story/Poem Generator, Social Media Content Generator, Financial Analysis, etc, the list goes on

Tools: **

HuggingFace: Try hosting your own LLM on your machine via HuggingFace vLLM, the inference may take a lot longer, and this will show you why GPUs are being used as opposed to CPU.
Rent a GPU: When your simple summarization takes 30 mins, try to rent a GPU (cheap ones are available all over the world), and host the LLM on the machine with GPUs available.
Amazon Bedrock: This could also be in the toying part. But if you have the ability, you can test out Bedrock and see how it enables you to use Foundational Models
Streamlit: Once you've built out a project and you're hitting the APIs, but you're still running from the terminal, you can try out building something like Streamlit to give you a quick UI. This will give you the ability to see how you can take input from an UI, pass it to the API, and show results back to the user.

Obviously, you can get much more complex than what I've provided here. This is just a starting point for you to kick the tires a bit with GenAI tools and projects. You'll be able to grasp some of these concepts a lot better after doing this.

Note: I am not hitting on anything like propagation, gradients, traditional NLP, or any serious academic concepts that exist underneath the hood of these GenAI tools. I come from Data Science, but they're not concepts I'm qualified to speak to, frankly.

Talking (and listening)

We love talking AI. But sometimes we don't need to do the talking. I like to ask people what they think of AI who aren't even in the Data or SWE spaces. I like to ask people who are deeply familiar with NLP what areas of traditional ML/NLP are relevant to understand in order to better grasp GenAI.

There's a time to talk and a time to listen. We should have discernment enough to know when each time is upon us. I frankly love to listen to really smart people talk about these topics. And yet sometimes, somehow, I'm the most knowledgeable about the situation in the room. I'll still happily admit I'm no expert, but have the context to speak to where it is and where we are going.

Conclusion

There are many ways to feed a cat (thanks Nicholas Pylipiw for the fun adaptation of this phrase). You can learn however you best see fit. These are some ideas depending on your level of commitment. These are also, in no way, exhaustive lists. You are at liberty to use your creative imagination to find the best way that works for you! The possibilities are nearly endless.

GenAI I can say, is not disappearing anytime soon. It will take many forms over the coming years, and it will continue to have impacts on our lives we can't really foresee at the moment. You can prepare yourself to be better informed to what these tools are, and what they aren't, by spending some time with them. Let's get out of our fairy tale domes and learn their capabilities as well as their limitations. We will be a better informed public, less at risk of causing global hysteria, by adequately informing ourselves to these tools.

Sharing the publication or leaving a comment is very valuable for me! What did you think? How are you using AI in your life or at work?

Dancing with Data

Discussion about this post