5.0free~5 min

Vagueness — 'make this better' and its cousins

Artifact: your vague phrase replaced with the dimension you actually care about

1. The failure mode

The cousin of make this better is any verb that hides the actual target:

  • Improve.
  • Polish.
  • Clean up.
  • Optimize.
  • Refine.
  • Tighten.

All of these mean change it in some direction I'll recognize when I see it. The model can't recognize what's in your head. It picks a direction. Sometimes the direction matches; mostly it doesn't.

The fix is naming the dimension. Better into shorter. Polish into more imperative voice. Optimize into fewer database round-trips. The named version is something the model can act on and you can verify.

2. The vague verb → specific dimension table

A starter map. The pattern: every vague verb expands into 3–5 specific dimensions. Pick the one you actually mean.

VagueSpecific dimensions
BetterShorter / clearer / more correct / more robust / faster
PolishTighten language / fix grammar / match house style / remove hedges
Clean upRemove dead code / extract helpers / consistent naming / consistent formatting
OptimizeReduce time complexity / reduce memory / reduce DB calls / parallelize
RefineMake scope smaller / make API narrower / remove edge cases / simplify happy path
TightenRemove redundancy / cut words / merge similar branches / cap loops

The point isn't to memorize the table. It's to notice the vague word in your own prompts and ask: which of these did I actually mean?

3. The fix in practice

Same code, two prompts.

textVibe — verb hides the target
Optimize this function.

The model picks. Sometimes the optimization is shorter code. Sometimes it's parallel processing. Sometimes it's caching. Most of the time it's not what you wanted.

textSpec — verb names the dimension
Reduce the number of database round-trips in this function. It currently makes one query per item in a loop (the N+1 pattern). Rewrite to use a single query.

Constraints: the existing function signature and return type must stay.

Now the verb (reduce) has a target (database round-trips) and a bound (one per item → one total). The model has a specific job. You can verify the result by counting queries.

4. Try it