---
title: "Prompt Engineering: Warum KI den Menschen braucht"
license: CC-BY-4.0
license_url: http://creativecommons.org/licenses/by/4.0
license_note: "The recording page https://media.ccc.de/v/clt25-271-prompt-engineering-warum-ki-den-menschen-braucht states: Licensed to the public under http://creativecommons.org/licenses/by/4.0. Confirmed 2026-10-18. The same sentence is in the event description returned by the public media.ccc.de API. No speech-subtitle track was listed. This note follows that description. It is an original summary, not a transcript."
source: https://media.ccc.de/v/clt25-271-prompt-engineering-warum-ki-den-menschen-braucht
source_license: "CC BY 4.0"
attribution: "Summary by iinventedlight (AI-assisted). Source: Kerstin Mende-Stief, Prompt Engineering: Warum KI den Menschen braucht, Chemnitzer Linux-Tage 2025 (CC BY 4.0)."
date: 2026-10-18
status: published
category: talks
tags: prompting, language models, training data
checked_on: 2026-10-18
sources:
  - id: S1
    title: "Prompt Engineering: Warum KI den Menschen braucht"
    authors: "Kerstin Mende-Stief"
    url: https://media.ccc.de/v/clt25-271-prompt-engineering-warum-ki-den-menschen-braucht
    accessed: 2026-10-18
    license: CC BY 4.0
    attribution: "Kerstin Mende-Stief, Prompt Engineering: Warum KI den Menschen braucht, Chemnitzer Linux-Tage 2025, CC BY 4.0, http://creativecommons.org/licenses/by/4.0, https://media.ccc.de/v/clt25-271-prompt-engineering-warum-ki-den-menschen-braucht"
---
# Prompt Engineering: Warum KI den Menschen braucht

License: [CC BY 4.0](http://creativecommons.org/licenses/by/4.0). Source: https://media.ccc.de/v/clt25-271-prompt-engineering-warum-ki-den-menschen-braucht

## Summary
Kerstin Mende-Stief, speaking at the Chemnitzer Linux-Tage on 22 March 2025, argues that a person should not trust an AI system blindly. The description says this holds whether the model is open or commercial, and whether it is built for general use or for one purpose. The quality of the result depends on two things: the training data, and the instructions. The web is described as the only source large enough, and cheap enough, to train language models, and as a source that also carries bias and unwanted content. Unclear instructions, or the wrong model, lead to inaccurate and faulty answers. Both the training and the use of the system need human work. The talk, as announced, covers how large language models work and how they are trained, the problems that come with that (including ethics, copyright, privacy, and hallucination), and prompt-engineering practices that aim at more precise and more relevant answers. This note follows the description on the recording page. The page lists no speech subtitles, so it does not report what was said on stage beyond that description.

## Why it is here
Askers want a prompt that gets a good answer. This talk's description says the wording of the instruction changes the result, and that a confident reply still needs a person, because the training data and the model choice can both be wrong.