Two Ways to Hit the Target

Opening Scene

An archer who wants to hit a target more reliably has two genuinely different options available. They can adjust their sights — a quick, reversible correction made shot to shot, requiring no change to the archer’s own trained body. Or they can retrain their stance and draw itself, a slower, deeper change that reshapes how every future shot is made, regardless of the sights. Getting a language model to do what you actually want has this same fundamental fork: adjust your prompt, or fine-tune the model.

In Plain English

Prompting shapes a model’s output by changing the instructions and context given to it at the moment of use, without altering the model’s underlying weights at all. Fine-tuning further trains a model’s actual weights on additional examples, changing its underlying behavior more permanently. Both are legitimate, widely used ways to steer a model toward a desired outcome, and this series exists specifically to build the judgment for choosing between them deliberately, rather than defaulting to one out of habit.

The Old Way

Before both prompting and fine-tuning existed as clearly distinct, well-understood levers, earlier approaches to customizing a system’s behavior looked meaningfully different:

Customizing a machine learning system’s behavior for a new task often meant training an entirely new, narrow model from scratch, without a large pretrained model to start from at all.
There wasn’t yet a lightweight, reversible way to steer a model’s behavior at the moment of use, the way prompting now allows.
The distinction between changing a model’s weights versus changing only its input context wasn’t yet a well-established, deliberate design choice for practitioners to make.

Both prompting and fine-tuning, as distinct, deliberately chosen levers, only became genuinely practical once large pretrained language models existed as a shared, adaptable starting point.

What’s Changing (and Why AI Is the Reason)

Large pretrained language models gave rise to prompting as a genuinely new, lightweight lever — a way to steer behavior without touching the model’s weights at all, connecting directly to this content library’s dedicated prompt engineering series.
Fine-tuning evolved from “the only way to specialize a model” into one deliberate option among several, chosen specifically when prompting alone proves insufficient.
Choosing between these two levers deliberately, rather than defaulting to either, has become a genuine practical skill this series builds article by article.

The Metaphor, Fully Extended

The Archer	Prompting vs. Fine-Tuning Concept
Adjusting the sights: quick, reversible, shot to shot	Prompting: changing instructions and context, not the model’s weights
Retraining the stance and draw: slower, deeper, more permanent	Fine-tuning: further training the model’s actual weights
Choosing which correction actually fits the problem	Choosing deliberately between prompting and fine-tuning for a given task
Both being legitimate ways to hit the target more reliably	Both being legitimate, widely used ways to steer a model’s output

For Beginners: What to Actually Do

Practice identifying, for a given model behavior problem, whether it’s plausibly fixable through better instructions alone, without touching the model itself.
Learn the basic distinction between changing a model’s input context (prompting) and changing its actual weights (fine-tuning).
Get comfortable with the idea that neither lever is inherently “better” — the right choice depends on the specific problem at hand.

For Practitioners and Leaders: The Deeper Layer

Build organizational fluency in both levers, since defaulting to only one regardless of the problem is a genuine, common inefficiency.
Recognize prompting and fine-tuning as points on a spectrum of customization effort and permanence, not a binary choice.
Use this series to develop a deliberate decision framework, built out fully by Article 18, for choosing between them on real projects.

Quick Recap

Prompting changes a model’s instructions and context at the moment of use, without altering its weights.
Fine-tuning further trains a model’s actual weights on additional examples, changing its behavior more permanently.
Both are legitimate, widely used levers, and choosing between them deliberately is a genuine practical skill.
This series builds that judgment article by article, from what each lever actually does through to a full decision framework.

Where This Fits in the Series

Article 1 introduced the fundamental fork between prompting and fine-tuning. Article 2 looks more closely at what adjusting the sights — prompting — actually changes about a model’s behavior.