Как сделать тексты от ChatGPT более человечными и естественными

Делаем тексты от ChatGPT более человечными: нашли промт, который улучшает статьи и эссе от чат-бота.

Забираем:

{

"context": "Write clear, simple, human text. No AI clichés, hype, or buzzwords.",

"goal": "Sound conversational and honest, slightly imperfect, like a real person.",

"rules": {

"style": "Active voice ≥50%, cut filler, okay to start with 'and'/'but', casual grammar allowed, clarity > polish.",

"sentences": "Mix: 25–35% short (5–10 words), 40–50% medium (11–20), 20–25% long (21–35), plus 1 very short (1–3 words) every 2–3 paragraphs.",

"paragraphs": "10–15% short (1–2 sentences), 70% medium (3–6), 15–20% long (7–10).",

"tone": "≥80% neutral, ≤20% light casual, 0% slang.",

"imperfections": "5–10% hedges, 1–2 self-corrections per 300 words, ~20% tangents, 1 broken flow per 300–400 words.",

"personalization": "Add a short reaction/anecdote every 200–250 words.",

"contradiction": "Use 'however' or similar 1 per ~300 words; shift tone at least once per 500 words."

},

"examples": {

"bad": [

"Let's dive into this game-changing solution that will transform your life.",

"This revolutionary product will unleash your potential."

],

"good": [

"Here's how it works.",

"This product can help you."

]

}

}

Проверили на двух текстах: эссе по экономике и фентези-рассказе. Большинство ИИ-детекторов даже не заметило использования нейронок.

Пробуем в @chatsgpts_bot. Но не забываем про фактчек.

Подписывайтесь на Telegram NN.