Comment mieux écrire avec l'IA

2025-05-04 · Source: Génération IA · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, extended

Summary

Benoît Raphaël, with AI engineer Thomas Mahier and expert AI Jeff, presents a method to improve AI-generated text quality, addressing its common predictability. The core issue is identified as a mathematical problem: language models maximize likelihood, choosing the most probable words, which minimizes information in Claude Shannon's sense of "surprise." This results in texts that are statistically average and lack distinctiveness. The proposed solution involves applying constraints, inspired by Georges Perec's Oulipo movement, to force AI out of its statistical comfort zone. This includes eliminating clichés, imposing specific narrative structures like the Wall Street Journal's "Kabob" method, and managing elements like rhythm, fluidity, and tension to create "open loops" that maintain reader attention. The article emphasizes that content and structure must precede style, and that reasoning models (e.g., GPT 5.1 Thinking, Claude Opus 4.5, Gemini 3) perform better due to their self-prompting capabilities.

Key takeaway

For AI Engineers and Prompt Engineers aiming to elevate AI-generated content beyond generic outputs, you should prioritize structured prompting that imposes specific constraints. Focus on defining content and narrative structure before stylistic elements, as this fundamentally alters the AI's information retrieval and expression. Experiment with "open loop" techniques and cliché removal to inject surprise and maintain reader engagement, moving beyond mere statistical probability to create truly informative and compelling text.

Key insights

AI-generated text predictability stems from models maximizing statistical likelihood, minimizing informational surprise.

Principles

• Information is surprise; predictability reduces information.
• Constraint liberates AI from statistical averages.
• Content and structure dictate style, not vice-versa.

Method

Improve AI text by applying constraints to eliminate clichés, impose narrative structures, and manage rhythm/tension, forcing models to select less probable, more informative tokens.

In practice

• Use "thinking" mode LLMs (e.g., Gemini 3, Claude Opus).
• Force AI to avoid cliché words and antithetical constructions.
• Implement narrative structures like "Zoom In, Zoom Out" for richer content.

Topics

• AI Writing
• Prompt Engineering
• Information Theory
• Narrative Structures
• Attention Engineering

Best for: Prompt Engineer, AI Engineer, AI Student

Related on AIssential

• Mastering AI Storytelling: Crafting Prompts for Captivating Narratives — Mastering AI storytelling prompts is a core competency for generating compelling narratives with generative AI.
• It’s not just one thing — it’s another thing — A specific sentence structure is now a strong indicator of AI-generated corporate communications.
• How the AI Writing Panic Is Making Us All Worse Writers — Stylistic evasion of AI writing patterns is a temporary, self-defeating strategy against advanced pattern-matching models.
• Your AI Writes Like a Robot Because Nobody Taught It to Listen — AI writing tools require measurement infrastructure to adapt voice, not just better prompts.
• My Make-Believe Project Hail Mary: Writing with Generative AI as Unknown Intelligence — Simulating "unknown intelligence" with generative AI reveals effective communication strategies.
• AI is making journalistic language more repetitive and predictable – and it’s a problem for all of us — AI's pervasive use in journalism risks linguistic homogenization and decline, particularly via "model collapse" from training on synthetic text.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Génération IA.