J'ai appris à Claude à ne plus rien oublier : installe ça en 5 minutes

J'ai appris à Claude à ne plus rien oublier : installe ça en 5 minutes

I Taught Claude Never to Forget Anything: Set This Up in 5 Minutes

Title history
  • J'ai appris à Claude à ne plus rien oublier : installe ça en 5 minutes — observed on 20261111-2335 UTC (YouTube)
🎙 ThomRichard 👥 6K 📅 October 20, 2026 ⏱ 10 min 👁 632 📄 tutorial 🧭 2026-11-11
Available in: English (current) Français

Keywords

decision journalClaudeChatGPTAI memoryproductivity

Summary

The video demonstrates how to implement a ‘decision journal’ for AI assistants, particularly Claude and ChatGPT, to ensure they remember, explain, and alert users about previous decisions. The creator introduces three key capabilities an AI should have: remembering decisions, explaining them, and alerting when a user is about to go against them. A concrete example of a customer service team is used to illustrate the consequences of an AI not respecting prior decisions, such as unauthorized refunds. The tutorial shows step-by-step installation in both Claude (via project instructions) and ChatGPT (via project instructions), using a single markdown file as the journal. The prompt instructs the AI to check the journal before proposing options and to flag contradictions. The demonstration shows that after logging a decision (e.g., ’never send emails directly to clients without validation’), both AIs correctly alert when the user tries to automate sending. The creator also briefly mentions his own AIOS (AI Operating System) integrating the decision journal across projects. The video concludes with three actions: maintaining a journal, explaining the why behind decisions, and configuring instructions to check the journal.

183 words

Critical Evaluation

Value of the Information & Strength of the Argument

The value lies in a simple, reproducible method to reduce wasted time and errors when using AI, illustrated with a realistic scenario. The argument is pragmatic: without a record, AI will re-suggest discarded options, leading to inefficiency and potential mistakes. The demonstration in both Claude and ChatGPT is concrete and easy to follow. However, the argument is based on personal experience and a single simulated case; it does not include quantitative data or comparative analysis, which limits its scientific weight.

Scientific Rigor, Source Quality, Title Accuracy

The video is a tutorial, not a scientific study. It does not reference external research or credible studies; the only sources are the creator’s own resources (a template and prompt) and the tools themselves. The title promises a quick fix, which is somewhat accurate but overstates the scope: it teaches a method to track decisions, not to make the AI ’never forget’ everything. No comments were provided for analysis.

164 words

Title / Content Match

The title accurately describes the goal of implementing a decision journal to prevent AI from ignoring prior decisions, though it oversimplifies the scope.

Quality & Reliability

5/10

The video provides a practical framework for managing AI decision memory, but relies on anecdotal evidence and personal demonstration. It does not cite external scientific sources, and the effectiveness is not empirically validated.

Chapters

Cited Sources

Contribution & Novelties

The video offers a straightforward, hands-on method to enhance AI assistants’ memory consistency by introducing a ‘decision journal’ – essentially a structured file that logs decisions, rationale, and alternatives. Its originality lies in presenting a practical, cross-platform solution (works with Claude, ChatGPT, and others) that directly addresses a common pain point of AI forgetting prior context. The approach is novel in its simplicity: a single markdown file combined with strategic prompts.

Pour aller plus loin :

  • Decision-making — Wikipedia article on decision-making processes, providing background on why tracking decisions matters.
  • Institutional memory — Concept relevant to storing knowledge and decisions across an organization or system, analogous to the journal.
  • AI alignment — The field focused on ensuring AI behaves in line with user intentions, which the decision journal partially addresses for individual decisions.

133 words

Radar Profile

The profile shows moderate scores on information quantity and quality, a low score on technical depth, and a slightly lower reliability score. This reflects a practical tutorial that is useful for beginners but lacks scientific rigor or deep technical analysis.

Reliability 4/10