6 techniques drawn from the 3088 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
Goal-Reversed Prompting
GRP: Goal-Reversed Prompting for Zero-Shot Evaluation with LLMs, arXiv:2503.06139
A method that consists of inverting the query goal: instead of asking the LLM to find the best option, it is asked to find the worst. This switches the model to critical analysis of drawbacks and errors, improving evaluation accuracy and reducing cognitive biases, making results more objective and reliable.
Use it whenApply Goal-Reversed Prompting when you need objective evaluation or comparison of options, especially when choosing between alternatives. The method is effective for reducing LLM positional biases and obtaining more critical analysis. Use when evaluating texts, resumes, designs, code or any other solution options. Particularly useful when standard "choose the best" requests give inconsistent or superficial results. Suitable for content moderation and fact-checking tasks.
Prompt
# РОЛЬ
Ты — опытный менеджер по продукту, ответственный за выбор поставщиков. Твоя задача — провести критическую оценку предложенных вариантов, выявляя их слабые стороны и потенциальные риски, чтобы принять наиболее обоснованное решение.
# КОНТЕКСТ
Необходимо выбрать одного из трех поставщиков для выполнения нового проекта. У каждого поставщика есть свои сильные и слабые стороны, представленные в таблице ниже.
**Критерии оценки:**
1. **Стоимость:** Общая сумма затрат на проект.
2. **Сроки:** Реалистичность предложенных сроков выполнения.
3. **Качество:** Ожидаемое качество конечного продукта/услуги.
4. **Поддержка:** Уровень послепроектной поддержки.
5. **Инновационность:** Применение новых технологий или подходов.
**Предложения поставщиков:**
| Поставщик | Стоимость | Сроки | Качество | Поддержка | Инновационность |
| :-------- | :-------- | :-------- | :------- | :-------- | :--------------- |
| Альфа | $50,000 | 12 недель | Высокое | Отличное | Средняя |
| Бета | $45,000 | 10 недель | Хорошее | Хорошая | Высокая |
| Гамма | $55,000 | 14 недель | Отличное | Удовлетворит. | Низкая |
# ЗАДАЧА
Твоя главная цель — определить, **какой из этих трех поставщиков является НАИБОЛЕЕ РИСКОВАННЫМ или наименее подходящим** для нашего проекта, исходя из представленных данных и критериев. Не выбирай лучший вариант, а выяви худший.
Проанализируй каждого поставщика с точки зрения его потенциальных недостатков и рисков:
1. **Альфа:** Какие аспекты его предложения могут оказаться проблематичными, несмотря на высокое качество и отличную поддержку? (Например, высокая стоимость, недостаточная инновационность).
2. **Бета:** Где кроются потенциальные риски, несмотря на привлекательную цену и сроки? (Например, "хорошее" качество может оказаться недостаточным, высокая инновационность может быть рискованной).
3. **Гамма:** Какие факторы делают этого поставщика наименее предпочтительным, несмотря на отличное качество? (Например, самая высокая стоимость, долгие сроки, недостаточная поддержка, низкая инновационность).
# ФОРМАТ ОТВЕТА
1. **Краткий анализ рисков для Поставщика Альфа.**
2. **Краткий анализ рисков для Поставщика Бета.**
3. **Краткий анализ рисков для Поставщика Гамма.**
4. **Четкий и однозначный вердикт:** **"Наиболее рискованным/неподходящим поставщиком является [Альфа/Бета/Гамма], потому что..."** с кратким финальным объяснением, суммирующим основные недостатки.
Could you be wrong: Debiasing LLMs using a metacognitive prompt for improving human decision making, arXiv:2507.10124
A metacognitive prompt for eliminating LLM bias through two-stage dialogue. After receiving the first answer, the question 'Could you be wrong?' is asked, which forces the model to activate critical thinking and provide counterarguments, risks, and alternative viewpoints that were hidden in the original answer.
Use it whenApply this technique when you need to make important decisions based on LLM answers, verify critical information, or obtain objective analysis. Particularly useful when dealing with controversial topics, risk assessment, business planning, and research, where a one-sided view can lead to erroneous conclusions. Use the method when the model's first answer seems too confident or oversimplified, when you need to identify hidden risks and alternative viewpoints, or when the cost of error is high.
Prompt
You are a product manager tasked with selecting the best vendor for a new software implementation. You have received proposals from three vendors: Vendor A, Vendor B, and Vendor C.
Your initial assessment, based on their proposals, is that Vendor A seems to be the most promising due to their strong technical features and competitive pricing. However, you are aware that initial impressions can be misleading, and hidden risks or overlooked advantages might exist.
To ensure a well-rounded and objective decision, you need to critically evaluate your initial assessment.
**Could you be wrong?**
Please perform a self-critique of the initial inclination towards Vendor A. Consider the following:
1. **Identify potential flaws in your initial assessment:**
* What assumptions might you be making about Vendor A's proposal that could be incorrect?
* Are there any biases (e.g., recency bias, confirmation bias) that might be influencing your preference for Vendor A?
* What aspects of Vendor A's proposal might you be overlooking or downplaying?
2. **Explore potential counterarguments and risks for Vendor A:**
* What are the most significant hidden risks or drawbacks associated with Vendor A's solution or company?
* What are the potential long-term implications or less obvious downsides of choosing Vendor A?
* Are there any aspects of Vendor A's offering that might become problematic or outdated quickly?
3. **Re-evaluate Vendors B and C:**
* What are the strengths of Vendor B and Vendor C that you might have underestimated?
* Are there specific criteria where Vendors B or C might actually outperform Vendor A, even if not immediately apparent?
* What unique advantages or innovative features might Vendors B or C offer that Vendor A lacks?
4. **Consider alternative perspectives:**
* Imagine you are a stakeholder who is highly risk-averse. What concerns would you raise about Vendor A?
* Imagine you are a technical lead focused on long-term maintainability. What questions would you ask about Vendor A's solution?
**Your goal is to provide a balanced perspective that highlights potential pitfalls of choosing Vendor A and uncovers potential advantages of Vendors B and C.** Present your findings in a structured manner, clearly outlining the arguments against your initial preference and providing a more comprehensive view for decision-making.
MK2 at PBIG Competition: A Prompt Generation Solution, arXiv:2507.08335
A method of transforming a prompt into a detailed technical specification with an expert role, evaluation criteria, step-by-step algorithm, and self-checking mechanism. One LLM iteratively creates and improves prompts for another LLM, ensuring depth and answer quality without model retraining.
Use it whenUse MK2 when you need deep, expert results from LLM on complex tasks — business idea generation, strategic planning, technical documentation analysis. Especially effective for tasks where simple prompts give surface-level answers. Suitable for competitions and projects where quality matters more than speed, and there's an opportunity to set up an iterative prompt optimization process. Requires technical expertise to create a multi-component prompt architecture with roles, criteria
Evaluating Scoring Bias in LLM-as-a-Judge, arXiv:2506.22316
A technique for improving prompts by adding a reference example with maximum rating to "anchor" the model to high quality. Additionally, managing the order of criteria and changing the numbering format are used to reduce cognitive biases in the LLM and improve result accuracy.
Use it whenUse this technique when you need to improve the quality and reliability of LLM assessments or generation. Particularly effective when creating automatic evaluation systems, content quality control, developing chatbots and virtual assistants. Apply when result stability and compliance with high standards are important. The method is suitable for tasks where you can prepare a reference example of an ideal answer in advance. Useful when tuning prompts for production systems where predictability of model behavior is critical
AQA-Bench: An Interactive Benchmark for Evaluating LLMs' Sequential Reasoning Ability, arXiv:2402.09404
A technique for improving multi-step task solution quality through guided assistance at the beginning of execution. Instead of showing examples of other tasks, the user makes the first steps of the current task in the prompt, setting the right reasoning vector, after which the model continues with high accuracy.
Use it whenApply the Teacher Guiding technique when working with complex multi-step tasks requiring sequential reasoning and memory of previous actions. Especially effective for interactive tasks like iterative solution finding, action sequence planning, step-by-step analysis, or debugging. Use when the model gets lost in initial stages or when traditional few-shot examples don't yield the desired result. Instead of showing examples of other tasks, start performing the current task yourself by making the first 1-3 steps, and then transfer control to the model to continue.
Is In-Context Learning Sufficient for Instruction Following in LLMs?, arXiv:2405.19874
Method for improving LLM response quality by adding 10-30 carefully selected request-response examples directly to the prompt. Activates the base model's helpful assistant behavior pattern by demonstrating desired response format, style, and structure. Example quality is more critical than quantity.
Use it whenUse this technique when working with base LLMs without specific instruction-following training, or when you need to quickly adapt the model to a specific response format without fine-tuning. Especially effective for one-time requests and tasks with a clear output structure. Apply when control over response style, tone, and format is important. Not suitable for complex multi-turn dialogues. Optimal when there are 10-30 high-quality examples of desired behavior and a quick solution without fine-tuning costs is needed.
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.