1977 → 2026

The Baby Language Model
in One Line of AWK

The smallest, simplest, and most educational language model ever created — written in a single line of AWK.

What is an AWK Baby LLM?

A "Baby Language Model" is a tiny statistical model that learns the most basic thing about language: which words appear most often.

This one-liner AWK script is the most minimal version of a unigram language model — the same core idea that powered early statistical language models before deep learning took over. 

awk '{
    gsub(/[^a-zA-Z0-9 ]/, " ");
    for(i=1;i<=NF;i++) {
        w = tolower($i);
        if(w) count[w]++
    }
} END { for(w in count) print count[w], w }' text.txt | sort -nr

How Does It Work?

🧹

Clean

Remove punctuation and convert to lowercase.

📊

Count

Build a frequency table using AWK’s powerful associative arrays.

📈

Rank

Sort by frequency — the fundamental idea behind statistical language modeling.

Try the Baby LLM Live

Baby LLM vs Modern LLMs

Baby LLM (AWK)

  • ✓ One line of code
  • ✓ Runs instantly on any Unix system
  • ✓ Zero training cost
  • ✓ Perfect for learning the core idea

Modern LLMs (Grok, GPT, etc.)

  • ⚡ Billions of parameters
  • ⚡ Trained on trillions of tokens
  • ⚡ Understands context and meaning
  • ⚡ Can generate fluent, coherent text

Learn More

Mastering AWK

AWK in the Age of AI

  • "Why Awk for AI?" (1997) — Classic discussion on using AWK for data work in AI
  • AWK is still widely used today for fast data preprocessing and cleaning before feeding data into machine learning pipelines.
  • Modern LLMs (like Grok) are excellent at generating high-quality AWK scripts on demand.