Uploaded December 2025 | Updated September 2026, 28 minutes ago

---

Day 10/42: What Is Alignment?

Yesterday, we talked about fine-tuning.

But good answers aren’t just accurate.

They need to be helpful, honest, and safe.

That’s the goal of alignment.

Alignment doesn’t add facts.

It shapes *how* a model responds.

It’s why ChatGPT refuses some requests.

Why tone matters.

Why safety exists at all.

Missed Day 9? Start there.

Tomorrow, we explain how humans teach models preferences: RLHF.

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

#Alignment #AIExplained #LLM #short