Best way to self-study in order to upskill?

Like Alex said, build something on your own. It's daunting to build something on your own, so start with a tutorial. Here's a 5-part algorithm to escape tutorial hell and self-study. Guaranteed to succeed:

1️⃣ Find a tutorial with good reviews and complete the project with the instructor. Type out every line, don't copy/paste code.

2️⃣ Now do the exact same project again, using the tutorial

3️⃣ Now do the exact same project, but only use the tutorial for every other step

4️⃣ Now do the exact same project, but don't use the tutorial at all

5️⃣ Now do the same project, but deviate from the tutorial in some way

It's tedious but it works 😃 Success criteria is that you build something on your own.

I put more thoughts here: How to come up an idea which matched on personal limited skills?

Build something.

The classic learning trap is falling into an endless cycle of consumption. You read, you watch, you listen, and nothing comes out of it. This is how education works: Some materials provide you the concepts, but it's up to you to apply the concepts and truly internalize them.

As an educational resource, Taro falls into this bucket. This is why I always end my courses with a "homework" section telling you how you can apply the learnings from the course the next week (or even the next day) into your actual life.

When it comes to what and how to build, there's 2 broad options:

1. Side projects
2. Open-source

For data & AI in particular, side projects (namely getting visibility on them) can be tricky. I think open-source might be better for you there, especially as you're a senior engineer who presumably has strong code reading ability. Check out this thread: "Suggestions for good open-source AI projects I can contribute to?"

There are some great answers already!

Building on Charlie's answer, I'd give serious thought to the practical applications of anything new you learn.

• Where might it be useful and why?
• Why doesn't anything else solve that problem (at all or with ease)?
• What design decisions in the new technology are central to the role it's expected to play?
• What might be some other alternative design choices?
• What are the downsides of the various design choices possible—those that were made as well as those that were not?

This is help you learn the technology faster as well as retain it. It's easy to forget the theory of something, even if you practice the code. It's much harder to forget it when you compile the new learning in your existing mental model.

Focusing on the practical applications will also enable you to put the technology to use when appropriate. Isn't that the whole point of learning something new?

You can try to brainstorm any workflows or processes that you do each day to see if you can incorporate AI to automate the process.

• If you like to cook, you can build a tool that scrapes recipes across different sites, passes the page through an LLM to remove all of the unnecessary steps. Maybe you can build something where you can take all of the recipes you want to make in a week, and the LLM can give you a shopping list.