Ever wondered what really makes large language models (e.g. ChatGPT) creative?
Surprisingly, it’s not as complicated as it seems – it’s a few smart techniques that happen right at the end, just before you see the answer.

That final “thin layer” – the one I like to call the “Creative Shell”– wraps around all the heavy computation. It’s where the magic of creativity starts.

🤖 This is part 0/5 my new mini series: The Creative Shell – What Makes LLMs “Creative”?

When we type something as simple as “The sky is…”, the model runs through familiar steps:
1️⃣ Our text gets tokenized and embedded (words → numbers)
2️⃣ The neural network processes it (the “heavy lifting”)
3️⃣ Then… the “Creative Shell” enters 🪄

Steps 1 and 2 are widely discussed, while Step 3 is often overlooked. Yet Step 3 is where:
→ Boring predictions become interesting responses
→ Deterministic math becomes creative output
→ The same model produces infinite variations

In this mini-series, I’ll crack open a real LLM and show what happens inside this creative shell … live, visual, and real.

Not a toy model … the actual thing.

Next episode from here