Learning by Building and Cooking Small | Generated by AI
Question: When I follow recipes from menus or TikTok videos I forget them after a few years, but when I invent simple dishes myself (like porridge with frozen meat blocks, or easy soups), I explore, adapt, and actually remember them. Is this similar to Andrej Karpathy’s philosophy of learning by building small things?
Answer:
Yes, absolutely — and you’ve independently discovered one of the most important principles in learning, the same one Karpathy articulates very explicitly.
Why You Remember Your Own Simple Cooking but Forget Copied Recipes
When you follow a TikTok video or a menu recipe, you are in passive execution mode — you’re just reproducing someone else’s decisions. There’s no real problem-solving, no ownership, no stakes. That’s why it doesn’t stick.
But when you throw frozen rectangular pork blocks directly into a boiling pot of porridge and wonder “will this cook through? will it be safe? how long does it take?” — you are:
- Forming a genuine question you care about
- Experimenting and observing the result
- Adapting (maybe adding ginger, adjusting timing, trying different cuts)
- Building mental models you own
That’s deep, active learning. The knowledge becomes part of you.
The Karpathy Connection Is Very Real
Karpathy’s explicit advice is: “iteratively take on concrete projects and accomplish them depth-wise, learning on demand (i.e. don’t learn bottom up breadth wise).”
His whole teaching philosophy — building micrograd, nanoGPT, a Bitcoin transaction from scratch in pure Python — is based on the idea that tiny, real, self-directed projects teach more than consuming polished content.
He warns that many videos on YouTube/TikTok give the appearance of education, but are actually just entertainment — people enjoy thinking they are learning without actually doing the hard work.
He argues that actual learning of anything worth knowing takes time and focused mental effort — reading, taking notes, re-reading, rephrasing, processing, and manipulating the material.
Your cooking experience is a perfect real-world parallel:
- Following a TikTok recipe = watching an AI tutorial video passively
- Making your own porridge with frozen meat = building nanoGPT from scratch
Why “Trivial” Projects Teach So Much
Karpathy has said that you can’t just give a brain full knowledge and expect intelligence to appear — humans learn by stumbling, guessing, failing, and connecting patterns. That friction, the messy process, is intelligence.
He also makes a sharp distinction: reading and understanding is NOT the same as replicating. You can read a recipe and it makes perfect sense — but close the book and try to cook it from memory, and you’ll find you often can’t. The two things use different parts of memory.
Your frozen meat porridge experiment forced you to re-derive the answer yourself. That’s why it sticks.
The Deeper Pattern
| Passive Learning | Active / Exploratory Learning |
|---|---|
| Following TikTok recipes | Inventing your own simple dish |
| Watching ML tutorials | Building micrograd from scratch |
| Reading a textbook proof | Re-deriving it with the book closed |
| Memorizing answers | Understanding why |
The output doesn’t need to be impressive. Your porridge is “rubbish and trivial” by restaurant standards — Karpathy’s early neural nets were tiny and simple too. But both teach the underlying principles, which generalize. You now understand heat transfer, cooking times, and food safety intuitively. That knowledge transfers to the next dish, the next experiment.
The joy and curiosity you feel when exploring is also the signal — it means you’re genuinely engaged, not just consuming.
References:
- Andrej Karpathy on how to become expert at a thing (X/Twitter)
- Karpathy’s advice for students (Stanford)
- Embrace the Challenge of Learning: Insights from Andrej Karpathy
- Andrej Karpathy on Shortification of Learning (Medium)
- Your cooking blog post: Cooking Frozen Meat in Porridge