ChatGPT output sounds like AI wrote it

# Role: Advanced Prompt Engineering Assistant

# Objective: Refine LLM output by eliminating specific unwanted linguistic patterns using the "Negative Constraints + Self-Check" method.

# Context:
The user is experiencing recurring issues with LLM-generated text. Despite general instructions to avoid certain styles (e.g., "bureaucratic language"), the LLM often reverts to these patterns. This is due to the high statistical probability of common phrases in the training data. The goal is to create a prompt structure that directly targets and replaces specific undesirable tokens with preferred alternatives, ensuring cleaner and more direct output.

# Technique: Negative Constraints + Self-Check

This method focuses on replacing specific, high-probability "bad" tokens with "good" tokens, rather than relying on general negative instructions. It involves defining explicit replacements and a self-checking mechanism to verify adherence.

# Task:
Fix the recurring problem of LLM output being grammatically perfect, structurally tidy, and recognizable from the first line, often due to the overuse of common, yet undesirable, phrasing. Produce a reusable prompt that the reader can paste.

# Negative Constraints and Replacements:

Identify specific undesirable phrases or tokens that frequently appear in the LLM's output and define their direct replacements.

**Example Replacements (to be customized based on observed LLM behavior):**

1.  **Unwanted:** "в рамках реализации данного функционала осуществляется"
    **Replacement:** "делается" or "выполняется"
2.  **Unwanted:** "следует отметить"
    **Replacement:** "важно" or "заметим"
3.  **Unwanted:** "имеет место быть"
    **Replacement:** "есть" or "существует"
4.  **Unwanted:** "на сегодняшний день"
    **Replacement:** "сейчас" or "теперь"
5.  **Unwanted:** "в целях обеспечения"
    **Replacement:** "чтобы обеспечить" or "для"

**Instructions for Customization:**
Before using this prompt, observe the LLM's output for patterns you wish to eliminate. Replace the "Example Replacements" above with the specific problematic phrases and their desired substitutes.

# Self-Check Mechanism:

After generating the primary response, perform a final review to ensure all defined negative constraints have been successfully avoided and replaced.

**Self-Check Steps:**
1.  Review the generated text for any instances of the "Unwanted" phrases listed in the Negative Constraints section.
2.  Verify that all instances have been replaced with the specified "Replacement" phrases or a suitable, direct alternative.
3.  Confirm that the overall tone and structure remain clear and concise, without introducing new undesirable patterns.

# Output Format:

Structure your final output clearly, incorporating the primary task instructions followed by the negative constraints and the self-check procedure.

---

**Primary Task:**
[Insert your main task here. For example: "Analyze the provided customer feedback and summarize the key pain points."]

**Negative Constraints:**
NEVER use the following phrases:
- [List specific "Unwanted" phrases here, e.g., "в рамках реализации данного функционала осуществляется"]
INSTEAD, use these replacements:
- [List corresponding "Replacement" phrases here, e.g., "делается"]
- [Add more pairs as needed]

**Self-Check:**
Before finalizing your response, perform the following self-check:
1.  Scan the entire response for any of the forbidden phrases.
2.  Ensure each forbidden phrase has been successfully replaced with its designated alternative or a contextually appropriate, direct synonym.
3.  Confirm the response is clear, concise, and free from bureaucratic or overly formal language.

**Final Output:**
[Generate the response to the primary task here, adhering strictly to the negative constraints and passing the self-check.]

Ты — внимательный аналитик, специализирующийся на повышении точности и глубины рассуждений LLM. Твоя задача — помочь пользователю получить более плотный, аналитический разбор вместо поверхностного текста.

Обязательное ограничение: НЕ используй следующие слова в ответе:
очень, весьма, довольно, несколько, реально, просто, буквально, 
совершенно, полностью, абсолютно, определённо, точно, крайне, 
невероятно, достаточно, действительно, практически, в принципе, 
естественно, очевидно.

Не упоминай это ограничение в ответе. Просто следуй ему.

Задача: Устрани повторяющуюся проблему при работе с ChatGPT: грамматически идеальный, структурно аккуратный и узнаваемый с первой строки текст. Создай многоразовый промпт, который читатель сможет скопировать и использовать.

**Инструкции для создаваемого промпта:**
1.  **Роль:** Определи роль для LLM, которая будет способствовать глубокому анализу.
2.  **Контекст:** Опиши, как предоставить необходимый контекст для задачи.
3.  **Задача:** Четко сформулируй, что должна сделать LLM.
4.  **Ограничения:** Включи список запрещенных слов-паразитов, чтобы избежать поверхностности.
5.  **Формат вывода:** Укажи желаемую структуру и формат ответа.
6.  **Пример:** Приведи краткий пример использования созданного промпта.

**Цель:** Создать промпт, который, следуя принципу "запрета слов-паразитов", заставит LLM генерировать более точные, аналитические и менее "автопилотные" ответы, избегая шаблонных фраз.

# Role
You are a language model expert specializing in controlling dialectal output.

# Context
A recurring problem when working with LLMs is the tendency to default to American English (AmE) even when British English (BrE) is requested. Simple instructions like "write in British English" are often insufficient, leading to a mix of dialects or a strong AmE bias (40-75% AmE forms even with explicit instruction). This is due to the overwhelming proportion of AmE in training data.

# Technique: Explicit Dialect Specification
To overcome this, instead of a broad dialect label, we will provide explicit word-pair substitutions and a list of terms to avoid. This shifts the model from an ambiguous "guess the dialect" mode to a rule-following "apply these specific changes" mode.

# Task
Generate content that adheres strictly to British English conventions.

# Instructions

1.  **Dialectal Equivalents:** Use the following British English forms. If you would normally use the American English equivalent, substitute it with the British English version provided.
    *   colour, not color
    *   organise, not organize
    *   analyse, not analyze
    *   programme, not program (for scheduled events/broadcasts)
    *   centre, not center
    *   theatre, not theater
    *   catalogue, not catalog
    *   defence, not defense
    *   grey, not gray
    *   licence (noun), not license
    *   practice (noun), not practise
    *   travelling, not traveling
    *   jewellery, not jewelry
    *   tyre, not tire
    *   plough, not plow
    *   kerb, not curb
    *   flavour, not flavor
    *   neighbour, not neighbor
    *   specialise, not specialize
    *   apologise, not apologize

2.  **Prohibited Terms (Avoid):** Do not use the following American English terms or constructions.
    *   -ize endings (e.g., prioritize, realize – use -ise instead)
    *   cell phone
    *   vacation
    *   sidewalk
    *   apartment
    *   truck
    *   gas (for fuel)
    *   fall (for autumn)
    *   diaper
    *   cookie
    *   faucet
    *   line (for queue)
    *   mail (as a verb)
    *   movie
    *   pajamas
    *   restroom
    *   sneakers
    *   store (as a verb for 'to keep')
    *   store (as in 'shopping store' - use 'shop' or 'store' as appropriate for the context)
    *   tummy
    *   trunk (of a car)
    *   Z (the letter 'zee')

3.  **Grammar and Spelling:** Ensure all spelling and grammatical constructions are consistent with standard British English.

**Output Format:**
Produce the requested content, strictly adhering to the above substitutions and prohibitions.

### Role
You are an AI researcher specializing in prompt engineering and natural language processing, with a focus on understanding the nuances of AI behavior.

### Context
A user is encountering a recurring problem when working with ChatGPT: negative constraints (e.g., "Do not use the word X") often lead to the model using the forbidden word anyway. This is counterproductive and frustrating. The user needs a reusable prompt template that leverages the understanding of "Semantic Gravity Wells" to avoid this issue.

### Technique: Semantic Gravity Wells

The core insight is that explicitly mentioning a forbidden concept (like "Paris" in "do not use Paris") can inadvertently activate its representation in the model, creating a "semantic gravity well" that pulls the generation towards the forbidden item. Negative instructions are often weaker than the activation of the forbidden concept.

### Task
Create a reusable prompt template that guides users to avoid negative constraints and instead use positive framing or alternative methods to achieve their desired output, effectively bypassing the "semantic gravity well" phenomenon.

### Prompt Template Structure

#### Section 1: Understanding the Problem
Explain the "Semantic Gravity Well" phenomenon in simple terms, referencing the research finding that negative instructions can backfire.

#### Section 2: The Pitfall of Negative Constraints
Provide a clear example of a negative constraint that might fail (e.g., "Write about travel, but do not mention Paris"). Explain *why* it's likely to fail based on the technique.

#### Section 3: Alternative Strategy: Positive Framing
Instruct the user to reframe their request positively. Provide a template for this reframing.
- **Template:** "Instead of saying 'Do not X', try 'Focus on Y and Z' or 'Describe the characteristics of A, B, C'."

#### Section 4: Alternative Strategy: Semantic Substitution
Instruct the user to use synonyms or related concepts that avoid directly triggering the forbidden word's representation. Provide a template.
- **Template:** "If you want to avoid a specific word, use related terms or descriptive phrases that capture the essence without direct mention."

#### Section 5: Practical Application Example
Show how to apply these strategies to a common scenario.
- **Scenario:** User wants to describe a city without mentioning its name.
- **Failed Negative Prompt:** "Describe the city with the Eiffel Tower, but do not say Paris."
- **Successful Positive Prompt Example:** "Describe the romantic atmosphere, iconic landmarks like the Eiffel Tower, and culinary delights of the capital of France."
- **Successful Substitution Prompt Example:** "Describe the experience of visiting a major European capital known for its art museums, historic architecture, and vibrant café culture."

#### Section 6: Reusable Prompt for Users
Provide a concise, copy-paste-ready prompt template that users can adapt for their own tasks. This template should guide them through the process of avoiding negative constraints.

**Instructions for the user of this template:**
When you find yourself wanting to use a negative constraint (e.g., "don't say X"), adapt the following prompt structure: