J'ai testé GPT 5.6 : comment le rendre 100 X meilleur ? Je te montre !

Value of the Information & Strength of the Argument

The video delivers valuable, practitioner-focused information on optimizing ChatGPT/Codex workflows. The concept of Loop Engineering—structuring prompts to create observation-decision-action-verification cycles—offers a credible methodology for improving AI reliability. The emphasis on generating an AGENTS.md file to centralize instructions and prevent file scatter is a practical tip likely to aid power users. The argumentation is mostly experience-based, with the creator drawing on personal usage and tests. He clearly explains reasoning limitations of LLMs and why explicit planning outperforms purely autonomous objectives, which is a sound argument. However, the value is somewhat diluted by constant self-promotion and a lack of external references or benchmarks to substantiate performance claims. The comparative analysis of token pricing and limits is concrete and useful, but presented assertively without official sources.

Scientific Rigor, Source Quality, Title Accuracy

The scientific rigor is moderate; the video offers practical know-how but not an evidence-based review. The only external link provided is to the creator’s own training platform, which serves as a source for his proprietary prompt systems. No scientific papers, official documentation, or independent benchmarks are cited, reducing the reliability of technical claims. The title’s promise of making GPT 5.6 ‘100 times better’ is clearly exaggerated and not substantiated by any measurement. Nonetheless, the content does deliver several actionable improvements, aligning partially with the title’s spirit. The video’s promotional nature affects its objectivity, but the core techniques (e.g., AGENTS.md, loop engineering) are plausible and align with common practices in the AI prompt engineering community.

The title promises a dramatic improvement ('100 times better'), but the content offers concrete configuration tips and techniques that plausibly enhance use, though the quantitative claim is hyperbole.

6/10

The video provides practical, actionable advice on using ChatGPT/Codex, but it is heavily promotional and relies on subjective comparisons rather than scientific evidence. Technical claims are not backed by external sources, and the emphasis on the creator's own paid training undermines objectivity.

Markers derived by PSI from the transcript: the creator did not define chapters.

• 00:00Introduction: promise of the video, presentation of the four skills (Loop Engineering, prompt system, MCP, skills).
• 04:30Explanation of the Codex architecture and the AGENTS.md file, use of /init function.
• 12:00Deep dive into Loop Engineering: observation, decision, action, verification loops, and the importance of safety barriers.
• 20:00Presentation of interface functions: plan, objective, fork, multitasking, and fan-out agents.
• 30:00Comparison of ChatGPT 5.6 and Claude (token limits, pricing, reset policies), argument that ChatGPT is more generous.
• 42:00Discussion of settings, auto-approval, and closing remarks; promotion of the paid training.

• Le Meilleur de l'IA (formation) — The creator promotes this paid training as the source for the complete prompt systems and updated resources mentioned in the video.