Uploaded January 2026 | Updated September 2026, 1 hour ago

---

Day 16/42: What Is Inference?

Yesterday, we covered reasoning.

Now we move to runtime.

Inference is the moment a trained model generates an answer.

When text appears word by word, you’re watching inference live.

The model predicts one token, adds it to the context, and repeats.

No planning.

No final draft hidden in advance.

This is why answers can change mid-sentence.

And why speed and cost matter.

Missed Day 15? Start there.

Tomorrow, we talk about waiting time: latency.

I’m Louis-François, PhD dropout, now CTO & co-founder at Towards AI. Follow me for tomorrow’s no-BS AI roundup 🚀

#Inference #LLM #AIExplained #short