09Artificial intelligence at The Mariscal

1. You're talking to an AI

Under art. 50 of Regulation (EU) 2024/1689 (the Artificial Intelligence Act), you have the right to know when you're interacting with an AI system. At The Mariscal, these are artificial intelligence:

• Mariscal AI in the Academy chat: a language model answers, using the context of the lesson or page you're on.
• The analyses of your exercises, games and exams, and your estimated level.
• The exercises and notebooks generated from an analysis or from a text.
• The analysis of images you upload (for example, reading the text in a photo).
• The automatic translation of comments (machine translation, not a conversational model).

Each of these features says so on screen. What is not real-time AI: the lessons, the videos, the book, the guides and the course exercises, which we make and check ourselves.

2. Content generated or edited with AI

• Some of the illustrations, the 3D models in the demos and the images on the sites were made with AI tools, and we checked and edited them ourselves.
• We write the texts ourselves; sometimes we use AI for drafts, translations or proofreading, and a person always checks them before they're published.
• We don't publish generated images, audio or video that imitate real people without their permission.
• The answers, analyses, exercises and notebooks Mariscal AI generates carry a visible label saying they're generated content.

3. What AI can and can't do

• It can get things wrong: explain something badly, make up a fact, a command or a feature that doesn't exist, or over-correct you. If in doubt, check the lesson, the official documentation or ask a person.
• It isn't a human teacher or a professional adviser.
• Don't use it for medical, legal, financial or safety decisions.
• The marks and the level it estimates are for guidance only: they don't lock you out of anything and have no effect outside the Academy.

4. What data it uses

To answer you, the AI receives your message and the minimum context it needs (the lesson or page, the point in the video, your results if you ask for an analysis). The models are run by Groq and Cerebras (USA), through the API and a server of The Mariscal's own (in Germany (EU)). We don't use your data to train models. All the details are in the Privacy policy.

5. What it costs

Mariscal AI features are paid for with credits: you see the cost before you use them. If a request fails because of a problem on our side, we give the credits back. The conditions are in the Terms and conditions.

6. AI in client projects

When we build an AI agent or feature for a client, the client decides what it's used for and is the one who deploys it to their users. The rules (human oversight, telling people it's an AI, what data it uses) are in the Terms of service.

7. Ask a person to review it

If an answer, a correction or an assessment seems wrong or unfair to you, write to hey@themariscal.com with the link or a screenshot: a person will review it and, if the AI got it wrong, we'll give you your credits back.