AI on the platform — what it may and may not do

用途#

Trust in AI arises from clear rules that can be looked up — and that are held in place in the code by guards, not only in a policy.

使用方法#

1. 1Labelled: presenter and voice-over are visibly marked as AI-generated above every slide, with “What exactly?” for the details (图 1).

   

   

   图 1 The notice above every slide: presenter and voice-over are AI-generated, with “What exactly?” — below it the slide it refers to.

2. 2Suggestion instead of decision: in the workshop nothing lands on a slide unasked; for papers there is a grade proposal, a person sets the grade.
3. 3Bound to sources: answers in the course pass through a verification chain; what is unsupported is marked as such; the inventory is checked against sources on a schedule.
4. 4Guardrails: attempts to get the AI out of its role, to query internals or to cause harm are fended off and logged.
5. 5Choice of model: every tenant decides which language models the workshop may use; one of them works in the EU (Paris).

设置与开关#

Where AI explicitly does not decide#

Labelling under the AI Act#

Every call is recorded#

提示#

• The checking tool for papers is explicitly classified in the guideline as a high-risk system within the meaning of the European AI Act.
• There is deliberately no AI score for cheating — and no machine that judges an incident in an exam.

亲自试用#

相关功能#

• Academic guidelineThe rules for academic work against which papers are checked — each with a number and legal basis, public and as a PDF.
• Fact check of the contentVerifiable statements in courses are checked against sources by two models — when new slides are generated and on a schedule across the whole inventory; every case is recorded in a file that cannot be retouched.
• Text models of the workshopWhich language model writes the slide texts can be enabled per tenant — three models from different providers, one of them processed in the EU (Paris); models that cannot be chosen are greyed out with a reason.
• AI guardrailsImmutable rules before every model call, a deterministic check of every input and a check of every output — with a log.
• Proctoring in examsCamera and behaviour monitoring during an online exam — what is recorded and why is stated openly beforehand; incidents are always reviewed by a person, never by a machine.
• Competence diagnosis (competence check)What can the learner already do, what is this assessment based on, what is still missing — and how does the course help in a targeted way? With tasks from practice, evidence per criterion and reassessment.