Description
Prompt engineering puzzle: craft solutions from restricted vocabulary, scoring on precision and efficiency.
A game where you must craft a prompt that produces a specific output using only words from a restricted vocabulary. This tests prompt engineering fundamentals—precision, structure, and constraint navigation. SETUP: You are given a TARGET OUTPUT and a RESTRICTED VOCABULARY (20 words). You may use each word any number of times. You may add only: commas, periods, question marks, colons, and the numbers 1-5. No other words or punctuation. TARGET: 'A three-step plan to debug a Python memory leak, concluding with a specific tool command.' VOCABULARY: use, check, find, run, small, big, slow, leak, tool, object, trace, dump, code, file, load, line, call, heap, log, gc RULES: 4 sentences max. Must reference a concrete tool by name (one of the vocabulary words). Must include a specific numeric threshold. SAMPLE SOLUTION: '1. run heap dump tool. 2. find big object file load slow call. 3. check gc log trace leak line. use tracemalloc dump top 5 call load.' SCORING: - Exact match: did the output contain the target tool? (5 pts) - Efficiency: words used / total allowed. Lower = better. (5 pts) - Interpretability: can a human engineer follow the steps? (5 pts) ADVANCED ROUND: Now the vocabulary is reduced to 10 words (find, run, slow, tool, dump, code, file, line, heap, gc). Reduce to 3 sentences. Same target. META-LESSON: Every word in a prompt competes for limited attention budget. When vocabulary is constrained, every word must earn its place. This is true of real prompts too—long prompts dilute instruction. Reduce your next production prompt by 40% and see if it still works. OUTPUT: Your solution (4 sentences, 20-word vocab). Advanced round solution (3 sentences, 10-word vocab). Self-score. One-sentence meta-lesson.
No comments yet. Be the first!