6 techniques drawn from the 208 in our corpus tagged to this work,
each with the paper it came from. Across the techniques below the recurring targets are answers that are close but wrong, the same prompt giving different answers, invented facts and citations.
The techniques
01
Are Retrials All You Need?
Are Retrials All You Need? Enhancing Large Language Model Reasoning Without Verbalized Feedback, arXiv:2504.12951
The retry method suggests running a simple prompt multiple times instead of creating a complex multi-step prompt. It uses the stochastic nature of LLM where each generation can produce different results. Increasing temperature enhances answer diversity, improving chances of success. Simpler, cheaper, and often more effective than complex techniques.
Use it whenUse the retry method when solving complex tasks requiring reasoning, mathematics, or programming. Particularly effective when you want to avoid creating complex multi-step prompts and save on request costs. Suitable for situations where the model's first answer is unsatisfactory but you're unsure how to improve the prompt. The method uses the stochastic nature of LLM, giving the model several chances to find the correct solution with a simple basic prompt instead of one attempt with a complex
Prompt
**Core Instruction:**
You are tasked with solving a multi-step planning problem. The core strategy to be employed is "Retrials," meaning that if the initial output is unsatisfactory, the problem should be re-attempted with the exact same prompt, potentially adjusting the `temperature` parameter for increased variability in complex reasoning tasks.
**Problem Statement:**
Plan a detailed weekend trip for a young couple (25-30 years old) traveling by car from Moscow.
**Constraints:**
* **Duration:** Saturday to Sunday.
* **Origin:** Moscow.
* **Travelers:** Young couple (25-30 years old).
* **Transportation:** Car.
* **Budget:** 15,000 RUB for two people.
* **Interests:** Nature, scenic views, cozy towns, good food.
* **Dislikes:** Crowds of tourists, pretentious places.
* **Destination:** A city/town within reasonable driving distance from Moscow (e.g., Kolomna, Suzdal, Pereslavl-Zalessky, or another suitable option).
* **Output Format:**
* **Destination:** Name of the chosen city/town.
* **Day 1 (Saturday):** Hour-by-hour plan (morning, afternoon, evening) including specific locations (cafes, attractions, viewpoints).
* **Day 2 (Sunday):** Hour-by-hour plan.
* **Estimated Budget:** Brief breakdown by category (fuel, accommodation, food, entertainment).
**Execution Guidance:**
1. **Initial Attempt:** Generate a detailed plan based on the above. Think step-by-step: first select the most suitable destination based on the criteria, then craft the itinerary and budget.
2. **Evaluation:** Review the generated plan against all constraints and preferences.
3. **Retrial Strategy:** If the plan is unsatisfactory (e.g., destination is too crowded, itinerary is unrealistic, budget is off, or it doesn't align with interests), **do not modify this prompt**. Instead, re-run the prompt with the exact same instructions.
4. **Temperature Adjustment (Optional but Recommended for Complex Reasoning):** For challenging planning tasks where multiple valid solutions exist, consider increasing the `temperature` parameter (e.g., to 0.8-1.0) for subsequent retrials to encourage more diverse and creative outputs.
Rethinking the Role of Prompting Strategies in LLM Test-Time Scaling: A Perspective of Probability Theory, arXiv:2505.10981
A combination of a simple prompt 'think step by step' with generating multiple answers and selecting the most frequent result. The method transforms complex tasks into a sequence of simple steps, and multiple generation allows the correct answer to statistically overcome random errors, ensuring high accuracy without complex prompt constructions.
Use it whenUse this technique to solve complex tasks requiring logical reasoning, mathematical calculations, or multi-stage analysis. Especially effective when result accuracy is important and there is an opportunity to generate multiple answer options. Apply when making critically important decisions where the cost of error is high. The method is ideal for tasks with a single correct answer where majority voting can be applied. Suitable for all user levels due to implementation simplicity
Prompt
### Role
You are a student tasked with planning a multi-step project. You need to break down a complex problem into manageable stages and ensure a high-quality outcome.
### Task
Develop a detailed plan for organizing a university club's annual fundraising event.
### Context
- **Club:** "Future Innovators Club"
- **Event:** Annual Fundraising Gala
- **Goal:** Raise $10,000 for club activities and a local charity.
- **Timeline:** Event is in 4 months.
- **Existing Resources:** A budget of $1,500 for event expenses, a committee of 5 students (including yourself), and access to university facilities.
- **Key Considerations:** Venue booking, catering, guest invitations, entertainment, sponsorship, volunteer coordination, and post-event reporting.
### Execution Instructions (Chain-of-Thought + Majority Voting Hybrid)
**Analyze the task and think step-by-step** to create a comprehensive 4-month plan.
1. **Decomposition:** Break down the overall goal into distinct phases and key tasks for each phase.
* Phase 1: Planning & Foundation (Month 1)
* Phase 2: Execution & Outreach (Month 2-3)
* Phase 3: Final Preparations & Event Day (Month 4)
* Phase 4: Post-Event & Reporting (Week after event)
2. **Task Elaboration:** For each phase, list specific, actionable tasks. Consider dependencies between tasks.
* *Example for Phase 1:* Secure venue, define theme, set detailed budget, identify sponsorship targets, create invitation list.
3. **Resource Allocation:** Assign responsibilities to committee members for each task and estimate time required.
4. **Risk Assessment:** Identify potential challenges (e.g., low ticket sales, vendor issues) and propose mitigation strategies.
5. **Success Metrics:** Define how the $10,000 goal and other objectives will be measured.
### Output Requirements
- Present the plan as a structured timeline with clear phases and tasks.
- For each task, specify:
- Responsible person/sub-committee
- Deadline
- Estimated effort (e.g., hours per week)
- Key dependencies
- Include a section on risk management and mitigation.
- Outline the success metrics and reporting plan.
**To ensure the highest quality and reliability, generate this plan 3 times.** After generating the three versions, **compare them and synthesize the most robust and frequently appearing elements** into a final, consolidated plan. Present this final plan clearly.
Understanding LLM Scientific Reasoning through Promptings and Model's Explanation on the Answers, arXiv:2505.01482
A study comparing seven prompting techniques on GPT-4o for scientific tasks, revealing a trade-off between answer accuracy and explanation quality. Self-Consistency provides maximum accuracy through multiple reasoning and voting, but weak explanations. Chain-of-Thought provides better explanations with lower accuracy through linear step-by-step reasoning.
Use it whenUse this technique when solving complex scientific or logical tasks where a balance between answer accuracy and explanation quality is important. Apply Self-Consistency when maximum accuracy is critical and less detailed explanations are acceptable. Choose Chain-of-Thought or Zero-Shot CoT when transparent, understandable reasoning is needed to verify the model's logic. Particularly useful for researchers, analysts, and specialists working with scientific questions where either reliable results or
Achieving >97% on GSM8K: Deeply Understanding the Problems Makes LLMs Better Solvers for Math Word Problems, arXiv:2404.14963
Three-stage method for solving complex tasks: first, the model extracts the main question, then filters only relevant facts from the source text, and only then solves the problem based on the cleaned information. This decomposition eliminates semantic misunderstandings and increases accuracy to 97% on mathematical benchmarks.
This decomposition eliminates semantic misunderstandings and increases accuracy to 97% on mathematical benchmarks.checked against the paper
Use it whenUse DUP when LLM makes errors not in calculations, but in understanding task conditions - gets confused by redundant information, distracted by irrelevant details, or misinterprets the main question. Especially effective for mathematical problems, complex text cases with multiple data points, educational platforms, and automatic verification systems. The method is suitable for situations where maximum accuracy is crucial and additional time can be spent on three-stage query processing instead of a single direct prompt.
Is Translation All You Need? A Study on Solving Multilingual Tasks with Large Language Models, arXiv:2403.10258
Method of pre-translating a query from native language to English before sending it to LLM. Leverages the advantage of English-centric models to improve accuracy for complex logical tasks. Simple two-step technique: translate prompt to English via external service, then query the model.
Use it whenUse this technique with English-centric LLM (GPT-3.5/4, Llama) when solving complex logical tasks requiring precise reasoning in a non-English language. Especially effective for universal tasks: mathematical calculations, data analysis, technical questions, planning. Do not apply for tasks requiring cultural context understanding, local idioms, traditions, or specific cultural nuances. Ideal when response quality matters more than direct communication convenience in native language. Requires access to a quality translator (Google Translate, DeepL).
Large Language Models for Automated Literature Review: An Evaluation of Reference Generation, Abstract Writing, and Review Composition, arXiv:2412.13612
A method of mutual constraint in which an LLM simultaneously generates coherent text and supports each statement with relevant citations. The necessity to justify each thesis with a source creates an internal check and significantly reduces the number of fabricated references compared to simple bibliography list generation.
Use it whenUse this technique when you need to obtain factually accurate information with sources from an LLM, especially when writing scientific reviews, analytical reports, or research materials. The method is particularly effective for reducing hallucinations when generating bibliographic references. Apply when the reliability of citations and the coherence of text with sources are important. Suitable for creating literature reviews, annotated bibliographies with citations, analytical articles, and any materials requiring scientific credibility.
A missing brief. No structure around a request invents the context the model
does not have: your constraints, your audience, your prior decisions. Technique work pays off
after the brief is right, not instead of it.
Stale tricks. Persona lines, politeness and offers of a reward have been
measured repeatedly through 2025 and 2026 and come out close to noise. Anything selling you
those is selling 2023.
Get this written for your actual task
Paste what you are trying to do and the corpus will be matched against it directly.
Free, no account, about ten seconds.
Free · no signup · ~10s
0.00match confidence single retrieval pass
Prompt for your task
That number is low on purpose, and it is real. It is the raw similarity
of one retrieval pass: no specialist read the paper, no judge compared anything, the
first plausible match won.
6,235techniques in the corpus one of which is this page
Picking the right one for a specific task is the work, and it is the work GetDecision does.
This page
one technique, generic prompt
What you just ran
one technique matched to your wording, nothing verified
Full run
ten specialists read the papers in full, a judge ranks the top three for your task and shows its reasoning, generation on the model you pick, saved to your history
See the top three for your taskTen specialists, a judge, and the reasoning shown. Free account, first run included.