Sights That Work for Any Target

Opening Scene

Adjustable sights work, in principle, on any target the archer might face — near or far, still or moving — without ever needing to be permanently reshaped for one specific case. That flexibility is a structural property of the sights themselves, not just a matter of convenience. Prompting carries this exact same structural flexibility relative to fine-tuning, and it’s worth understanding as a genuine advantage in its own right, not simply “the cheaper option.”

In Plain English

Because prompting doesn’t change a model’s underlying weights, the same underlying model remains available for a wide range of different tasks, simply by changing the prompt. A fine-tuned model, by contrast, has been shaped specifically toward its training data’s patterns, and that specialization — while valuable for its intended task — makes it structurally less suited to different tasks without further fine-tuning. This tradeoff between prompting’s generality and fine-tuning’s specialization is fundamental, not incidental.

The Old Way

Before this generality-versus-specialization tradeoff was well understood as fundamental, it was sometimes framed as simply a matter of cost:

Prompting’s flexibility was sometimes framed only in terms of lower upfront cost, understating its genuine structural advantage of supporting many tasks from one unchanged model.
Fine-tuning’s specialization was sometimes treated as a strictly positive property, without full acknowledgment of the generality it necessarily trades away.
There wasn’t yet a well-established framework for weighing generality against specialization as a fundamental tradeoff, independent of cost considerations alone.

Recognizing this as a fundamental generality-versus-specialization tradeoff, not just a cost difference, gives a fuller, more accurate picture of what each approach genuinely offers.

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

Teams increasingly value prompting’s ability to support many different tasks from a single, unchanged model, especially where task requirements are still evolving or genuinely varied.
This connects directly to the multi-task flexibility many production systems now require, where maintaining a separate fine-tuned model per task would introduce significant real overhead.
This tradeoff feeds directly into the decision framework built out fully in Article 18 — task variety and stability are among the most concrete factors it weighs.

The Metaphor, Fully Extended

The Archer	Prompting’s Generality Advantage
Adjustable sights working, in principle, on any target faced	Prompting supporting many different tasks from one unchanged model
Flexibility as a structural property, not just a cost saving	Generality as a structural property of leaving the model’s weights unchanged
A specialized stance being less suited to unfamiliar targets	A fine-tuned model being less suited to tasks outside its specialization
Choosing flexibility deliberately for varied or evolving conditions	Choosing prompting deliberately for varied or evolving task requirements

For Beginners: What to Actually Do

Practice identifying, for a given project, whether the task requirements are likely to stay stable or continue evolving over time.
Learn to frame prompting’s flexibility as a structural advantage, not just a lower-cost option, when explaining the tradeoff to others.
Get comfortable recognizing that fine-tuning’s specialization is a genuine tradeoff, not a strictly superior property.

For Practitioners and Leaders: The Deeper Layer

Weigh task variety and evolution explicitly when choosing between prompting and fine-tuning, not just upfront and ongoing cost.
Consider the real overhead of maintaining separate fine-tuned models across multiple distinct tasks, versus one flexible, prompted model.
Feed this generality-versus-specialization tradeoff directly into the decision framework built out fully in Article 18.

Quick Recap

Prompting’s flexibility is a structural advantage: the same unchanged model supports many different tasks.
Fine-tuning’s specialization is valuable for its intended task, but trades away that generality.
This is a fundamental tradeoff, not simply a matter of which approach costs less upfront.
Task variety and stability are concrete, weighable factors in choosing between the two approaches.

Where This Fits in the Series

Article 13 covered prompting’s structural flexibility advantage. Article 14 looks at combining both approaches together, rather than treating them as mutually exclusive.