Stop Planning Your AI Learning. Start Here Instead

The Question I Couldn’t Answer

Found it interesting. I’m wondering what the best learning system for me is. I don’t have the patience to sit there and do entire courses like you do. I like YouTube. There’s so much to learn and I don’t know where to start.

My colleague sent me this after reading my post about technical literacy.

He got it. He knows he needs to learn. And I left him with nothing.

I wrote 2,000 words convincing him there’s a problem, then told him to “build his own learning path.” That’s not advice. That’s outsourcing the hard part back to him.

So here’s the actual answer to “where do I start?” based on what I’ve learned struggling through Ed Donner’s AI course and breaking things in my own projects.

TL;DR: Three learning patterns below based on how your brain actually works. Pick the one that sounds least painful. Do the first step today.

Why “So Much to Learn” Keeps You From Learning Anything

Here’s the trap: You Google “learn AI” and get 47,000 courses, all promising to be comprehensive. You open YouTube and there are 10,000 videos on “AI fundamentals for beginners.”

More resources doesn’t help. It paralyzes.

The second problem: Every learning path assumes you’re a university student with time and patience. You’re not. You’re grabbing 30 minutes between meetings with a brain that’s already full from your actual job.

Most advice treats learning like a curriculum problem. It’s not. It’s a “how does my specific brain actually absorb new information” problem.

Three Patterns I’ve Seen Actually Work

Not “learning styles” in the educational psychology sense. Just three patterns I’ve watched people (including myself) use to go from stuck to learning.

Pick the one that sounds least painful. That’s your path.

Pattern 1: The YouTube Grazer (This is My Colleague)

No patience for courses. Needs variety. Gets bored if something goes too long.

The mistake most people make: Treating YouTube like a research project. Watching “AI for Beginners” videos until your eyes glaze over, never actually trying anything.

What actually works: Deliberate grazing with a simple rule - watch, then do.

Week 1 - Just watch three things (total time: ~45 minutes):

1. Andrej Karpathy “Intro to Large Language Models” - first 20 minutes only

   • Stop at the 20-minute mark even if you want to keep watching

   • This gives you the mental model without overwhelming you

2. 3Blue1Brown “But what is a neural network?” - full video, 19 minutes

   • This one is worth watching completely

   • Best visual explanation of how neural networks actually work

3. Two Minute Papers - pick any recent AI breakthrough that sounds cool (3-5 minutes)

   • Just browse their recent uploads and pick one that catches your eye

   • Gets you current on what’s possible

After watching - The critical step most people skip:

Open ChatGPT or Claude. Ask it: “I just watched [video topic]. Explain it back to me in simpler terms.”

Then ask: “What’s one thing I could try right now to see this in action?”

Do that thing. Even if it takes 5 minutes. This turns passive watching into active learning.

Week 2 - Follow whatever grabbed you:

Did neural networks seem interesting? Watch 3Blue1Brown’s series on them.

Did the LLM stuff feel relevant to your work? Search “how do LLMs work” and watch 3-4 different people explain it. Notice where they agree and where they differ.

Something specific catch your attention? Go down that hole.

Week 3 - Pick one thing to try yourself:

By now you’ve seen enough to know what interests you. Pick one small thing:

• Build a simple ChatGPT prompt that solves a real problem for you

• Try making a custom GPT for something you do repeatedly

• Use Claude Projects to help with a work task

You’re building a mental map, not taking a test.

After a month, you’ll naturally know what you want to go deeper on. The grazing phase gave you enough context to know what questions to ask.

You’re doing it wrong if: You’re still watching “AI basics for beginners” videos in week four. That’s procrastination wearing a learning costume. Pick something specific and try it.

YouTube channels that work for grazers:

• Andrej Karpathy (deep but clear)

• 3Blue1Brown (visual explanations)

• Two Minute Papers (what’s new and possible)

• AI Explained (current developments explained simply)

• Computerphile (technical depth made accessible)

Pattern 2: The Problem-First Person (This is Me)

I can’t learn abstract concepts. I need a real problem or I get bored and quit. Even small problems work.

Pick one annoying task from your actual work. Use AI to solve it. Learn only what you need for that.

Examples that worked for me:

• Trying to get Claude to process a Jira CSV export (failed for some time before it worked; learned about API limitations)

• Creating permanent notes for my Obsidian vault (prompting with guidelines and a getting a structured markdown template as output)

• Building a personal assistant that works with every available LLM model (in the early stage; base functionality is working; experimenting with agents.md and spec-driven development)

Start with this:

Pick your smallest annoying task. Tomorrow, open ChatGPT or Claude. Paste in your problem. See what happens.

Don’t research first. Don’t plan. Just try to solve the actual problem.

Day 2-3: Your first attempt probably didn’t work perfectly. Google “better ChatGPT prompts” or “prompt engineering basics” and read 2-3 articles. Try your problem again with what you learned.

Week 2: Search YouTube for “ChatGPT custom instructions” or “Claude Projects setup” - about 10 minutes total. Set it up for your specific use case.

Week 3: Is there a way to make this automatic or repeatable? Look into:

• Custom GPTs (if using ChatGPT)

• Claude Projects with custom instructions

• API access if you’re technical enough

What you’re actually learning:

• Prompt engineering (most immediately useful AI skill)

• How to evaluate AI output quality

• When AI helps vs. when it creates more work

• Your own comfort level with AI assistance

But you’re learning it by doing a real thing, not studying abstractions.

When to level up: When your problem is solved and you’re annoyed by the next one. Solve that. Repeat. Each problem teaches you something new.

Good first problems to try:

• Format messy meeting notes

• Draft responses to common email types

• Summarize long documents

• Generate first drafts of repetitive writing

• Analyze data you have in a spreadsheet

Pattern 3: The Structured Person

Maybe you actually like courses. You just need them to be time-boxed and goal-oriented.

Not “comprehensive.” Not “start your AI journey.” Just: learn one specific thing, finish it, use it for a few weeks, then decide what’s next.

Good focused courses (1-6 hours each):

1. “ChatGPT Prompt Engineering for Developers” (DeepLearning.AI, free)

   • If you want to use AI tools better right now

   • Taught by Andrew Ng and OpenAI team

   • Very practical, minimal theory

2. “AI for Everyone” (Andrew Ng, Coursera)

   • If you need the big picture first

   • Non-technical overview

   • Helps you understand what’s possible

3. “Building Systems with the ChatGPT API” (DeepLearning.AI)

   • If you want to build something

   • More technical but still accessible

   • Gets you from user to builder

The commitment: Pick one. Finish it. Then stop and actually use what you learned for 2-4 weeks before starting another.

The goal isn’t course completion. It’s: complete one course → use it in reality → let that experience show you what to learn next.

You’re doing it wrong if: Three courses in progress, zero finished. That’s collecting, not learning.

After you finish: Spend at least as much time using what you learned as you spent in the course. Real experience reveals what you actually need to learn next.

What You Do Right Now

Stop reading. Which pattern sounds least painful?

YouTube Grazer: Open YouTube. Search “Andrej Karpathy LLM”. Watch 20 minutes. Then open ChatGPT and ask it to explain what you learned.

Problem-First: Write down three annoying work tasks. Pick the smallest. Open ChatGPT or Claude right now. Paste it in. See what happens.

Structured: Go to DeepLearning.AI. Create free account. Start “ChatGPT Prompt Engineering for Developers.”

Don’t research other options. Don’t make a plan. Don’t wait until Monday.

Just do the next physical action.

What Actually Happens When You Start

After 4-6 weeks on any of these paths, you’ll hit something you don’t understand.

That gap is your next learning goal.

Not “learn AI.” Not “understand machine learning.” But specific things like “why isn’t this prompt working” or “what does temperature actually do.”

Currently I’m spending some spare-time building a personal assistant that grows with me on this learning journey. That’s not in any course syllabus. Figuring out how things work teach me more about how LLMs. work than watching ten more fundamental videos.

Technical literacy isn’t built by comprehensive learning. It’s built by accumulating specific understandings, one frustration at a time.

My colleague asked “where do I start?”

Start wherever the smallest friction is. Let that friction reveal the next thing to learn.

The path emerges by walking, not by planning.

---

P.S. This post is me answering the question I failed to answer in my previous article. If you’re reading this and thinking “but what about my specific situation?” – that’s the feedback I need. Reply and tell me what I’m still missing.

---