Yintoni i-Re-Prompting ifumaneka (and Why You'll Use It More Than You Think) The "classic" LLM ukuxhaswa i-one-and-done: 
 
 I-You type: "Write me an article about sustainability." I-model isibophelela kunye ne-essay ye-school generic. I-You type: "Write me an article about sustainability." I-model isibophelela kunye ne-essay ye-school generic. Yintoni akuyona ingxaki model. Kuyinto a . spec problem (I-re-emitting a refined prompt emva kokufumana i-first output) i-bridge esebenzayo kusuka ku: Re-prompting 
 
 Intention umntu → machine-readable instruction → usable output Yintoni unako ukufikelela kwi-"there is output" kwi-"these output is correct, structured, and shippable". I-Re-Prompting, i-defined I-Practice ye Emva koko, uqhagamshelane ngokugqithisileyo. Re-prompting adjusting the prompt’s content, structure, or constraints after inspecting the model’s first response Umzekelo: It is Umgangatho wokuqala ayikho "umgangatho". It is . feedback-driven telemetry Umzekelo omncinci "Nceda uqhagamshelane into malunga ne-Climate Change." Prompt v1: inkcazelo generic, akukho ingqungquthela, akukho ukubaluleka. Output: "Nceda uqhagamshelane i-explainer ye-350-450 iingoma Ukucaciswa kwe-climate change. Oku kuquka: Prompt v2: UK secondary school students 
 
 iingxaki ezimbini, 2) iingxaki ezimbini, 3) iingxaki ezimbini zonyango ziyafumaneka, kunye 4) ekugqibeleni kunye neengxaki yeengxaki ezimbini. Iimveliso efanayo. Ukusabela ezahlukeneyo. Ngenxa yokuba i-spec iye yenzelwe. Yintoni i-Single-Prompting ifumaneka (A Re-Prompting ayikho) Iingcebiso ezininzi ziquka ukuba unokufuneka ngokupheleleyo into efunyenwe ngexesha elandelayo. Oku kwakhona ngexesha ... ngexesha ke. 
 
 
 
 
 
 
 
 
 
 
 Dimension 
 Single Prompt 
 Re-Prompting 
 
 
 
 
 Interaction 
 One-way 
 Feedback loop 
 
 
 
 
 Info base 
 Only initial intent 
 Intent + output defects 
 
 
 
 
 Goal 
 “Get an answer” 
 “Get the right answer” 
 
 
 
 
 Best for 
 Simple asks 
 Fine-grained or high-stakes output Ukusebenzisana Ukucinga I-Feedback Loop INFO Base I-Intent yokuqala kuphela Intent + Imiphumo Imiphumo Ukucinga “Ukufumana impendulo” “Ukufumana impendulo elungileyo” Best ukuba Ukulungiselela Simple I-Fine-grained okanye i-high-stakes yi-output I-Real Value: Yintoni i-Re-Prompting ibekwe 1) Ukunciphisa ukufikelela Ngaba ungenza ukuba umlawuli wokuqala. Ungenza ukuba ungenza . iterate 2) I-Dynamic Calibration yeengxaki ze-complex I-brand voice, i-marketing ye-product, i-writing ye-technical, i-policy summaries - le nto I-Re-Prompting inikezela ukulungiselela i-tone kunye ne-structure ngokukhawuleza. moving targets 3) I-output engaphezulu I-temperature, i-sampling, kunye ne-model randomness inokukhawuleza i-drift. Ukongeza (“no slang”, “usebenzisa iziguli”, “kuquka iingcebiso zokusetyenziswa”) ukunciphisa i-variance. constraints Xa Uya Kufuneka Ukulungiselela? A Practical Trigger List Zonke imibuzo ayidinga umzila wokuqala. Kodwa lezi zibonise: 1) I-Output ifumaneka kwi-Core Goal inqwelana malunga neengxaki kodwa unxibelelana neengxaki yakho. Signal: "Wabhalisa iingcebiso zokuthengisa iiyure yokuhamba ebonakalayo ebonakalayo Ukusetyenziswa kwe-UK hikers. " Example ask: 72-hour battery iingxaki malunga ne-esthetics kunye ne-band material; i-battery ingxaki. Bad output: Ngaba ungenza i-"battery + hiking context" engabonakaliwe. Why you re-prompt: 2) Umgangatho ungaphendula Ukuba unemibuzo ye-machine-usable, ufumane i-prose. Signal: "Yenza le specs ye-telephone kwi-Markdown table kunye ne-headers Ngathi Example ask: Model | RAM | Storage  a numbered list. Bad output: ifomati kufuneka yi-explicit, kwaye ngokuvamile zisetyenziswa . Why you re-prompt: mini example 3) kukho iingxaki ze-logic okanye i-concept imibuzo embalwa, iingxaki, okanye "ngokufuneka kakuhle" kodwa ayikho. Signal: "Ukucacisa i-Chain-of-Thought yokukhuthaza nge-example yeemathematics." Example ask: I-COT ibizwa ngokuba yi-“give the answer directly” kwaye usebenzisa Yintoni “umzekelo”. Bad output: 2+3=5 inkcazelo yakho ayinxibelelana ne-definition, kwaye awukwazi ukunika isampuli elungileyo. Why you re-prompt: 4) I-too short okanye i-too long Ukuba unemibuzo iingcebiso, unamathele iingcebiso, okanye unamathele iingcebiso, unamathele iingcebiso. Signal: “Detailed” kunye “short” ayikho imiyalezo. Ziziphi vibes. Ukuguqulwa vibes nge . Why you re-prompt: structure and word limits I-5-Step Loop: Wabelane Re-Prompting kwi-system Nazi i-framework evimbini iteration kuya kuba thrashing ngempumelelo. Isinyathelo 1: Ukubala i-output (ngokusebenzisa i-checklist ye-3 axis) Nceda usebenzise le ezintathu ezininzi ngexesha: 
 
 
 
 
 
 
 
 
 
 Axis 
 What you’re checking 
 Example failure 
 
 
 
 
 Accuracy 
 Does it solve the actual problem? 
 Deletes the wrong rows when asked to handle missing values 
 
 
 
 
 Format 
 Is it structured as requested? 
 No code comments / no table / wrong schema 
 
 
 
 
 Completeness 
 Did it cover every must-have point? 
 Handles missing values but ignores outliers Ukucaciswa Ngaba kuthetha inkinga yokwenene? Ukukhusela izilwanyana ezahlukileyo xa kubhalwe ukulawula iimpawu ezikhoyo Ukucinga Yintoni i-structure efanelekileyo njenge-requested? No code comments / no table / isisombululo Ukucaciswa Yintoni iingxaki ziyi-must-have zonke iingxaki? Handles missing values but ignores outliers I-Pro-Tip: Wabelane i-"list of defects" ephakeme njenge-QA engineer: . Goal Drift / Format Break / Missing Points / Wrong Concepts / Verbosity Isinyathelo 2: Trace umonakalo back to the prompt Izixhobo ezininzi ze-output ziyafumaneka kwiimeko ezininzi ezinokufumana. 
 
 
 
 
 
 
 
 
 
 
 Defect 
 Likely prompt cause 
 Example 
 
 
 
 
 Goal drift 
 vague priorities 
 didn’t state the   selling point primary 
 
 
 
 
 Format break 
 format not explicit 
 “organise this” ≠ “Markdown table with headers…” 
 
 
 
 
 Logic error 
 wrong / missing definition 
 didn’t anchor the concept and example 
 
 
 
 
 Too brief 
 no structure requirement 
 didn’t specify steps, length, sections Ukucinga Imibuzo emangalisayo Ukukhangisa i Ukuhweba Point Ukucinga Ukucaciswa Umgangatho Non-Explicit “Yenza oku” ≠ “Markdown table with headers...” I-Logic Error Ukugqithisa / Ukugqithisa Ukungabonakali i-concept kunye ne-example Ukucinga akukho imfuneko yokwakha Musa ukucacisa iingxaki, ubude, iingxaki Ukulungiselela ukucinga: Ukuphendula i-Spec don’t blame the model first Step 3: Apply one of 4 high-leverage adjustments I-Strategy A: Ukongeza i-context + i-constraints (ukuguqulwa kwe-vagueness / i-goal drift) Ukongeza: i-audience, i-scenario, i-priority order, i-content yasekelwe, iiphakamiso eziluncedo. “Skrolela i-eco article.” "Wabhalisa iingoma ze-400 kubafundi ze-UK Year 9, inkqubo ye-recycling yeeskole, kuquka iingoma ze-3 ezisebenzayo, umnqweno we-friendly, akukho i-jargon, lokugqibela kunye ne-call to action." Before: After: Strategy B: Make format executable (ukuguqula i-format break) Qhagamshelana: uhlobo + scheme + isibonelo. After prompt snippet: Return a Markdown table exactly like this:
| Model | RAM | Storage |
|---|---|---|
| Example | 8GB | 128GB |
 Strategy C: Ukongeza okanye ukuguqulwa izibonelo (ukuguqulwa iingxaki) Ukuba umzekelo umzekelo umzekelo umzekelo, . show it the pattern Example for Chain-of-Thought (correct pattern): 
 
 
 
 I-Problem: "Sam unayo i-apples ezini, yenza i-2, yenza i-3..." Ukuphendula: 5−2 = 3, 3+3 = 6 Ukuphendula: 6 I-Strategy D: Ukulawula iinkcukacha kunye ne-structure + limits (ukuguqulwa kwe-verbosity) Ukucacisa iinkcukacha, iingcebiso zeengcebiso, iingcebiso zeengcebiso, kunye ne-"finished" efanayo. "Ukucacisa iShoot Few-Prompting ku Zonke iingxaki: UMAX Ngathi Prompt snippet: 3 steps 1 sentence 20 words Isinyathelo 4: Validate (kwixesha lokufumana ukuba uqhagamshelane kwakhona) Re-run the prompt. Re-score ngokufanayo 3 axes. Stop when: 
 
 
 
 Izixhobo zangaphakathi ziyafumaneka, ifomati efanelekileyo, Akukho iingxaki ezininzi zonyango. Ngaba ukhangela “umgangatho.” Chase . usable I-3-5 iterations. Ukuba unemibuzo emva kwe-5, ingxaki ingabonakaliweyo okanye ungenza iimodeli / izixhobo ezahlukileyo. Rule of thumb: Isinyathelo 5: Template win (ngokuthi awukwazi ukuqala ukususela ukususela) Xa isebenza ngokufanelekileyo, ukutya njenge a : prompt template 
 
 
 Iindawo eziqhelekileyo: imiyalelo, isakhiwo, imiyalelo yokufaka Iimpawu ze-variable: iimpawu ze-fillable ezifana ne-{audience}, ne-{constraints}, ne-{inputs} Example: Python data-cleaning prompt template Generate Python code that meets this spec:
​
1) Goal: {goal}
2) Input: {input_description}
3) Requirements:
   - Use pandas
   - Handle missing values using {method}
   - Handle outliers using {outlier_rule}
4) Output rules:
   - Include comments for each step
   - End with 1–2 lines on how to run the script
   - Wrap the code in a Markdown code block
 Njengoko i-re-prompting compounds: ukwakha i-library ye-prompts ezisebenza njenge-tools. A Full Iteration Walkthrough Thina ukuvelisa into efanelekileyo: Ukusetyenziswa njenge-operator 3 promo posts for a bubble tea brand Round 0: Prompt that fails “Write 3 Posts Social for I-Sweet Sprout Bubble Tea.” Prompt: i-repeatable, i-blind, akukho i-hooks, akukho i-platform iimpazamo. Output: Ukucaciswa 
 
 
 
 Ukucaciswa: Vague Brand Fit Umgangatho: akukho hashtags, akukho CTA Ukuphumla: 3 post, kodwa zero ukwahlukanisa I-Round 1: Ukongeza iimpawu ze-brand + i-platform format Re-prompt: 
 
 
 
 
 
 Brand: i-low-sugar, i-0-cal iinketho I-Taro Boba Milk Tea I-Shop Vibe: I-Instagrammable Interior I-Style: i-energic, i-emojis, ekupheleni kunye ne-2-3 i-hashtags Differentiation: zonke iingxelo zihlanganisa (1) ukutya, (2) iifoto, (3) low-cal , kodwa nangokungabikho "ukwenza into ngoku". Output improves Indawo 2: Faka i-action trigger (ukufaka + i-promo) Ukongeza ku: 
 
 
 
 indawo: Manchester City Center Ukukhangisa: BOGOF ukuya 31 Oct I-CTA: “kufika emva kokusebenza / ngexesha lokugqibela” Ngoku i-output iya kuba usetyenziswa. Iimpawu eziqhelekileyo (and how not to waste your life) Isisombululo 1: Ukuguqulwa yonke into ngexesha elandelayo Uyakwazi ukufumana into efanelekileyo ukuba uqhagamshelane yonke inkxaso ngexesha elinye. Ngokuba iteration. one major variable I-Error 2: "I-Model is Bad" njenge-diagnosis yakho yokufakelwa Kwiminyaka emininzi, i-prompt yakho ayikwazanga: 
 
 
 
 
 
 Ukucaciswa Ukucinga Ukucaciswa Umzekelo Imibuzo yokuphumelela Mistake 3: Infinite iteration chasing perfection Qinisekisa umgaqo yokukhipha. Ukuba ilungileyo kwaye usebenzayo, utshintshe. Isisombululo se-4: Ukunciphisa umyalezo wokuqala Ukuba isebenze isicelo, isampula kunokuba kunzima kakhulu kunokuba isibuyekezo esikhulu esifanayo. Isisombululo se-5: Ukubhalisa-ukubhalisa i-prompt efanayo kwi-platforms Yintoni okufanayo kwi-Hacker News iya kuthetha kwi-TikTok. Ngexesha elide. platform constraints I-Tooling ne-Workflow Tips 
 
 
 
 
 Yenza i "defect → fix" cheat sheet (i-format missing → add schema + example; ukuguqulwa → ukulawula iingongoma ezahlukileyo; i-concept error → add definition + exemplar). Ukubuyekeza kwi-3-5 outputs ngaphambi kokuba ukuguqulwa kwe-50. I-Store Prompts kwi-malet system (i-Notion, i-Obsidian, okanye i-git repo). Ukuba uqhagamshelane neqela le-team: uqhagamshelane neengxelo ze-prompt njenge-code. Ukucinga Re-prompting is a trick. It is the workflow. Ukuba umyalezo wokubhalisa iimfuno, — usebenzisa i-output ye-model njenge-logs yakho yeengxaki. Xa uye uye uyenze ngokwenene, uye uye uye uye "ukutshintsha iingxaki ze-ChatGPT" kwaye uye uqala ukuvelisa Uyakwazi ukufumana i-ship. re-prompting is debugging those requirements reliable text-and-code generators

This story contains AI-generated text. The author has used AI either for research, to generate outlines, or write the text itself. 

Read My Stories

Le audio iveliswe ngolwimi lokuqala lwebali!

Re-Prompting: I-Loop Eyenza I-Output ye-Meh ye-LLM kwi-results ezisetyenziswayo

About Author

IZIMVO

ZIJONGE IIMPAWU

ELI NQAKU LINIKEZELWE KU

Related Stories

Light-Mode

Classic

Newspaper

Minty

Dark-Mode

Neon Noir

Minty

HN StartUps