One model, three finishes

Same underlying weights. What gets done to them after determines whether it completes your sentence, answers your question, or answers it in one very specific voice.

Base, instruct, and fine-tuned aren't three different models — they're the same one, worked further each step.

One prompt, three outputs

Watch what changes — and what doesn't

Base

Raw block

No instinct to answer — just to continue whatever this text looks like.

Instruct-tuned

Sanded & painted

Learned that a prompt is a request, and a reply is what comes next.

Fine-tuned

Custom-engraved

Same helpfulness, forced into one exact voice, every single time.

Hand-written to illustrate the pattern each stage is known for — not captured from a live model.

So what

Fine-tuning changes how it always sounds. It was never the door for teaching it new facts — that's page 10's door 2.