hAIgopala › AI terms › Rapid Fire (3 terms per short) › Day 16

Instruction Tuning

Generative AI & LLMsChapter 4: Choosing the LLMTerm 75 of 253

Fine-tuning on instruction→response pairs so the model follows directions.

At Sunrise School

Gopal asks

Why does Bit follow “reply in 3 bullet points”?

Bit answers

Instruction tuning: training on thousands of instruction-and-answer pairs, so I do what you ask instead of just continuing the text.

In the real world

Teaching it to answer, not just continue text.

Watch the short

Short coming soonSeason 5 · Day 16Follow @haigopala

Quick check

Quick check. Does quality usually improve with more data and compute? Yes or no?

Taught together with

Constitutional AI / RLAIF · Scaling Laws

Practise it

Try this term and all 253 others in the free Model Training Arena.

Open the Arena →

Follow Bit daily: YouTube English · తెలుగు · हिंदी  |  Instagram English · తెలుగు · हिंदी