Learn AI
◆ TIER 1 · ORIENT

Fine-tuning —
when to reach for it

Most application teams never fine-tune anything — and ship great products anyway. Knowing when it actually pays (and when it's an expensive detour) is the senior skill. That's all this page installs.

01 · THE ESCALATION LADDER

Four rungs, cheapest first. Tap a rung to see what it buys — and what it costs.

Exhaust the cheap rungs first. What pushes you up a rung should be a failing eval — not vibes.

02 · WHAT IT CHANGES vs WHAT IT DOESN'T

Tap a card to flip it. Two honest columns.

✓ IT CHANGES
✗ IT DOESN'T

The classic line, worth memorizing: fine-tune for behavior, RAG for knowledge.

03 · THE METHODS, AT INTUITION LEVEL

Tap to flip. One line each — the intuition, zero math.

04 · SHOULD I FINE-TUNE?

Pick the symptom you actually have. Get the boring, correct answer.

MY SYMPTOM
THE CALL
Try structured outputs first
Constrained decoding / JSON mode fixes most drift for free. If schema discipline still fails under an eval, a small SFT run locks it in.
fine-tuning is a power tool — most days, the screwdriver wins · last reviewed 2026-06
The Recall DeckFoundations