Theory: Instructional prompting means giving the model a direct, explicit, well-scoped task description — the prompting equivalent of a clearly written problem statement on a problem set. The key engineering principle here is specification completeness: just as an underspecified control system has ambiguous behavior, an underspecified prompt has ambiguous (and often lower-quality) output. Good instructional prompts typically specify:
- The task (what to compute, derive, or explain)
- The format (equation, table, step-by-step derivation, code)
- The constraints (assumptions, units, precision, notation conventions)
- The audience/level (undergraduate vs. graduate, which changes rigor and vocabulary)
EXAMPLE 1
Prompt
Explain, what a decision tree is in machine learning.
Instructions:
Use simple language.
Keep the answer under 100 words.
Assume that audience are fresh graduates.
Here, the prompt gives instructions, like :
- Explain in simple language.
- Keep the answer under 100 words.
EXAMPLE 2
Weak prompt (underspecified):
Solve this circuit. A series RLC circuit has R = 100 Ω, L = 50 mH, and C = 20 μF, driven by a 10 V sinusoidal source.
Strong instructional prompt:
A series RLC circuit has R = 100 Ω, L = 50 mH, and C = 20 μF, driven by a 10 V sinusoidal source. Do the following, in order:
- Compute the resonant angular frequency ω₀.
- Compute the quality factor Q.
- Determine whether the circuit is underdamped, critically damped, or overdamped, and justify it mathematically.
- Present all final results in a table with units.
Assume ideal components and steady-state sinusoidal analysis. Show all formulas used before substituting values. The audience is 2nd year graduates with strong math background in Calculus and Algebra.