Startseite > Kalender-Artikel > BP: KI für Autoren (I)

BP: KI für Autoren (I)

Mittwoch 15. April 2026, von Dr. Wolf Siegert

I.

DJV Berlin - JVBB | Workshop
"Künstliche Intelligenz - Einführung für Journalist:innen" 15.04.26 (ehemals: 25.03.26) - 11 bis 17 Uhr - Markgrafenstr. 15, 10969 Berlin

Liebe Mitglieder,

Künstliche Intelligenz (KI) ist im Newsroom angekommen. Doch welche Tools sind Spielerei und welche echte Arbeitserleichterung?
Dieser Workshop bietet einen Einstieg in die KI. Wir klären die Grundlagen und zeigen, wie Sie KI effizient in Ihren beruflichen Alltag integrieren.

Zielgruppe:
Journalist:innen und Medienschaffende ohne oder mit geringen KI-Vorkenntnissen.
Bitte bringen Sie Ihren Laptop (oder Pad) mit.

Inhalt:
In diesem Workshop legen wir das Fundament für Ihre Arbeit mit Künstlicher Intelligenz.
Wir klären, welche Tools 2026 den Standard setzen und wie Sie diese effizient nutzen:

• KI-Landschaft 2026: Überblick über Sprach- und Multimodale Modelle (ChatGPT, Claude, Gemini, Perplexity).
• Stärken & Schwächen: Wo glänzt die KI (Zusammenfassungen, Strukturierung), wo stößt sie an Grenzen (Halluzinationen, Quellentreue)?
• Kostenfrei vs. Premium: Welche Tools bieten den besten Mehrwert ohne Budget?
• Grundlagen des Promptings: Einführung in das „Journalistische Prompting“ für Texte, Bilder und einfache Datenauswertungen.
• Tool-Check: Welche Anwendung eignet sich am besten für welchen Workflow?

Dozent:
Alexander Czekalla arbeitet als freier Dozent und Berater u.a. für den DJV und ist Reporter mit Schwerpunkt Online und Social Media bei der Märkischen Oderzeitung.

Termin:
Mittwoch, 15.04.26 (ehemals: 25. März 2026), 11:00 bis 17:00 Uhr

Teilnahmegebühr:
99 Euro für Mitglieder, 299 Euro für Nichtmitglieder

II.

In der ersten Übung geht es um die Nutzung von NotebookLM.

Die eigene Frage ist, ob man diese Anwendung auch ohne Cloud-Zugriff on premise , das heisst n u r auf dem eigenen Rechner nutzen kann. Um das herauszufinden, wird die Seite https://notebooklm.google/app geöffnet. Ein Zugriff ist aber nur möglich, wenn man ein Google-Konto anlegt oder das bestehende einsetzt. Das Ergebnis sieht dann, nach der Aktivierung der 2-FAKTOR-Authentifizierung ...

... so aus.

Gleich beim ersten Zugriff werden die kostenpflichtigen Angebote in den Vordergrund gestellt:

Will man das nicht, kann man sich auch per YouTube coachen lassen

Die Frage nach den Downloads auf eigene Endgeräte wird wie folgt beantwortet:

Zu guter Letzt wird nochmals in das System dieser Simple-Prompt formuliert:
"Kann ich notebooklm auf dem Rechner einsetzen?"

Wie nutze ich NotebookLM auf dem Rechner?

Browser öffnen: Rufen Sie notebooklm.google.com in Ihrem Browser (z. B. Chrome, Edge, Firefox) auf.
Anmelden: Melden Sie sich mit Ihrem Google-Konto an.
Loslegen: Erstellen Sie ein neues „Notebook“ und laden Sie Ihre Quellen (PDFs, Google Docs, Websites, YouTube-Links) hoch.

Wichtige Hinweise zur Nutzung

Keine Installation nötig: NotebookLM läuft direkt im Browser, es ist keine Software-Installation erforderlich.
Datenhoheit: Die hochgeladenen Quellen bleiben privat und werden laut Google nicht zum Training der öffentlichen KI-Modelle verwendet.
Funktionen: Sie können Quellen zusammenfassen, Fragen dazu stellen, Quizze erstellen oder Audio-Übersichten (Podcasts) generieren.
Synchronisation: Ihre Notebooks werden mit Ihrem Google-Account synchronisiert, sodass Sie auch vom Smartphone aus (über den Browser oder die App) darauf zugreifen können.
Grenzen: Es gibt Limits (max. 50 Quellen pro Notebook, 500.000 Wörter) und eine tägliche Begrenzung für Chat-Anfragen.

Es gibt zudem für Android und iOS eine offizielle App, die sich über den Google-Account mit der Desktop-Version synchronisiert

Und was lernen wir daraus? notebooklm kann nicht on premise, also als lokale Anwendung ausserhalb des Google-Imperiums eingesetzt werden.

III.

In der nächsten Runde wurde diese Anfrage via app.creamai gestellt: " what is the content of https://www.daybyday.press/article9514.html"

Die Antwort von GPT-4 Omni lautet:

The webpage at https://www.daybyday.press/article9514.html contains an article titled "BP: KI für Autoren" by Dr. Wolf Siegert, published on April 15, 2026. It discusses a workshop organized by DJV Berlin - JVBB, titled "Künstliche Intelligenz - Einführung für Journalist:innen," aimed at providing journalists and media professionals an introduction to the use of Artificial Intelligence (AI) in their work.
Key Details:

Workshop Overview:
Date and Time: April 15, 2026, from 11:00 AM to 5:00 PM.
Location: Markgrafenstr. 15, 10969 Berlin.
Target Audience: Journalists and media professionals with little or no prior knowledge of AI.

Workshop Content:
Introduction to AI tools: Discussion of leading AI models in 2026, including ChatGPT, Claude, Gemini, and Perplexity.
Strengths and Weaknesses: Evaluating AI capabilities in summarization and structuring, as well as limitations like hallucinations and source accuracy.
Cost-Effectiveness: Comparison of free and premium tools.
Prompting Techniques: Basics of "journalistic prompting" for generating text, visuals, and analyzing data.
Tool Assessment: Identifying the best tools for specific workflows.

Instructor:
Alexander Czekalla: A freelance lecturer and consultant, also a reporter specializing in online and social media for the Märkische Oderzeitung.

Participation Fee:
€99 for DJV members.
€299 for non-members.

Additional Notes:

The article also mentions the tool NotebookLM, an AI application by Google, and explores whether it can be used locally ("on premise") without relying on cloud services. The conclusion is that NotebookLM cannot function as a standalone local application and requires a Google account for browser-based use.
Features of NotebookLM include summarizing sources, creating quizzes, and generating audio overviews, with synchronization across devices via a Google account.

Overall, the workshop aims to empower journalists to effectively integrate AI into their workflows while addressing practical challenges and limitations.

Die gleiche Anfrage bei der nicht-kostenpflichtigen Version von Perplexity eingestellt, führt zu folgendem Ergebnis:

IV.

Frage an Claude: Kann ich Claude auf dem eigenen Rechner on premise nutzen?
o

Kurze Antwort

Nein, Claude kann derzeit nicht vollständig on-premise (lokal auf eigener Hardware) betrieben werden. Anthropic bietet die großen Claude-Modelle nicht zum lokalen Download oder Self-Hosting an.

Welche Optionen gibt es?
1. Anthropic API (Cloud)
Zugriff über die offizielle API
Daten werden an Anthropic-Server gesendet
api.anthropic.com

2. Cloud-Partner
Amazon Bedrock (AWS)
Google Cloud Vertex AI
Hier läuft Claude in deiner Cloud-Umgebung, was Compliance erleichtern kann

3. Alternativen für On-Premise

Wenn du unbedingt ein LLM lokal betreiben willst, gibt es Open-Source-Modelle:

LLaMA / LLaMA 3 (Meta)
Mistral / Mixtral
Qwen
Tools wie Ollama, LM Studio oder vLLM machen lokales Hosting einfach

Warum nicht on-premise?
Grund Erklärung
Geschäftsmodell Anthropic bietet Claude als Service an
Modellgröße Claude erfordert erhebliche GPU-Ressourcen
Sicherheit Anthropic behält Kontrolle über Safety-Maßnahmen

Wenn Datenschutz das Hauptanliegen ist:

Die AWS Bedrock-Option bietet u. a. Datenverarbeitung in der EU (Frankfurt) und vertraglich geregelten Datenschutz
Anthropic gibt an, API-Daten nicht zum Training zu verwenden (je nach Nutzungsbedingungen)

Aber Claude bietet eine on premise wirksame Software an:

IV.

TEST: Transcription einer Audio-Datei:

Ich sehe, dass du eine MP3-Datei hochgeladen hast! Lass mich die Audiodatei lesen und transkribieren.Checking the uploaded fileChecking the uploaded fileMP3-Dateien sind leider nicht direkt über den Computer transkribierbar, da ich keinen Netzwerkzugriff auf externe Transkriptions-APIs habe. Ich kann jedoch die Audiodatei über die Anthropic API in einem Artifact transkribieren lassen – Claude unterstützt nämlich Audioeingaben!Read the MP3 file as base64Read the MP3 file as base64Die Datei ist etwa 16 MB groß. Leider ist sie zu groß, um sie direkt über die API als base64 zu übertragen (Limit liegt bei ca. 5 MB für inline-Audio).
Hier sind deine Optionen:
Option 1 – Kürzere Aufnahme: Wenn du eine kürzere/kleinere MP3-Datei hast (unter 5 MB), kann ich sie direkt transkribieren.
Option 2 – Externe Tools (kostenlos):

Whisper von OpenAI – sehr genaue Transkription, auch auf Deutsch
otter.ai – einfach zu bedienen
Google Docs – Extras → Diktierfunktion (für Echtzeit)

Option 3 – Lokal installieren:
Falls du Python hast, kannst du Whisper lokal installieren:
bashpip install openai-whisper
whisper deine_datei.mp3 —language German
Möchtest du eine der oben genannten Optionen ausprobieren, oder hast du eventuell eine kürzere Audiodatei? Sie haben bis 20:00 keine kostenlosen Nachrichten./quote>

V.

Ausserhalb des Protokolls ein Hinweis auf das besondere Interesse an den Claude-Protokollen von Michael Dobler auf DrWeb vom 13. April 2026: Verändert Claude Mythos die Cybersicherheit?

Claude Mythos Preview ist kein spezialisiertes Sicherheitstool. Anthropic hat ein universelles Sprachmodell gebaut, das Code lesen, analysieren und auf Schwachstellen prüfen kann. Der entscheidende Unterschied zu früheren Modellen: Mythos verkettet mehrere Schwachstellen eigenständig zu vollständigen Angriffsketten. Im Linux-Kernel kombinierte das Modell separate Lücken zu einer kompletten Rechteeskalation, vom normalen Benutzer bis zum Root-Zugang.

Und in Ergänzung dazu der Beitrag Tomislav Bezmalinović auf heise.de vom 15. April 2026: Nach Anthropic Mythos: OpenAI kündigt GPT-5.4-Cyber an

Nachdem Anthropic mit dem KI-Modell Mythos in den vergangenen Tagen bis in höchste US-Regierungskreise für Aufregung gesorgt hat, stellt OpenAI nun ein eigenes, auf defensive Sicherheitsaufgaben zugeschnittenes Modell vor: GPT-5.4-Cyber.

VI.

Aus den im Verlauf der Veranstaltung geführten Gesprächen werden für die Folgetag BP: KI für Autoren (II) drei der Aufzeichnungen ausgekoppelt


VIELEN DANK ...
... für die vielen und vielfältigen Rückmeldungen auf diesen Beitrag.
Es wird daher sowohl an dieser Stelle als auch an den nachfolgenden beiden Tagen
 BP: KI für Autoren (II)
 BP & WS: KI für Autoren (III)
je zwei nachfolgende Einträge geben:

Journalismus unter Druck #4 – KI als Retter des Journalismus?. Mitschnitt einer Liveübertragung vom 21. Mai 2024 von der Heinrich-Böll-Stiftung auf Youtube:

Der Einsatz von Künstlicher Intelligenz im Journalismus birgt Risiken, hat aber auch Potential. Einerseits besteht die Gefahr, dass Arbeitsplätze wegfallen könnten, weil die KI Aufgaben
übernimmt, die bislang Journalist*innen ausgeführt haben. Auch die Qualität des Journalismus könnte leiden, wenn durch die Automatisierung Fehler oder Fake News verbreitet werden.
Gleichzeitig lassen sich – gerade im Lokaljournalismus – schwindende Ressourcen mit Hilfe von Technologie kompensieren.
Welche Regeln brauchen wir, um das Beste aus der KI rauszuholen und Risiken zu minimieren?
Mit:
Dr. Stephan Weichert, VOCER Institut für digitale Resilienz
Sascha Devigne, Chefredakteur STUDIO 47
Marie Todeskino, Mitglied der Chefredaktion bei Ippen.Digital
Ole Reißmann, Leiter Künstliche Intelligenz in der SPIEGEL-Gruppe
Moderation: Vera Linß, Medienjournalistin

.

Und hier als pars pro toto dieser Claude-generierte Text zum Thema:

50 THESES FOR A SCIENTIFIC REFORMATION
Artificial Dimensional Intelligence (ADI) applied to Thinking

A Preface

In 1517, Martin Luther walked to the door of a church in Wittenberg and nailed 95 theses to the wood. He was not trying to destroy the Church. He was trying to make it honest.
Science today occupies the position the Church held then: the unquestioned authority on what is real, what is true, and how the world is to be understood.
Like the Church before it, science has accumulated assumptions so old and so embedded that they are no longer examined. They are simply inherited.
These theses are not an attack on science. They are written in its spirit.
They begin where Einstein left off - with his quiet admission that the laws of mathematics, when they are certain, do not refer to reality; and when they refer to reality, they are not certain. That gap has never been closed. ADI is an attempt to close it.
I. On Reality and Measurement
1. Reality is what physics investigates. How we measure reality determines what we believe it to be.
2. Every measurement is a comparison. Every comparison requires a shared unit. Every unit carries hidden assumptions about the nature of the world.
3. When those assumptions go unexamined, the measurements appear precise while the understanding remains shallow.
4. The international system of measurement - the SI - defines seven fundamental units. Each unit belongs to a different dimension of reality. Those differences matter.
5. Mixing dimensional levels in a single expression produces mathematics that is internally consistent and physically misleading at the same time. This happens constantly, and largely without acknowledgment.
6. A reformation of measurement does not require new instruments. It requires the honesty to ask what our existing instruments actually capture - and what they silently discard.

II. On Dimension

7. The word dimension means different things in geometry, in physics, in mathematics, and in computing. This is not a minor terminological inconvenience. It is the source of deep confusion at the foundations of science.
8. Linear measurement - one direction, one axis - is the simplest form of quantification. It is also the most limited. The world is not a line.
9. Spatial measurement adds volume, weight, presence. It is three-dimensional in the most literal sense: it occupies space.
10. Temporal measurement adds change, movement, becoming. Time is not a fourth line added to three axes. It is a different quality of reality altogether.
11. When these three qualities - linear, spatial, temporal - are treated as interchangeable, geometry disappears from physics. What remains is calculation without shape.
12. Recovering dimension means recovering shape. And recovering shape means recovering the curvature that has been quietly removed from our models of the world.

III. On Time

13. Time, as we measure and imagine it, is a line. It has a beginning - the Big Bang - and moves in a single direction toward an end not yet determined.
14. This model was not discovered. It was chosen. Joseph Priestley (1733 - 1804) gave us the timeline in the 18th century, and we have lived inside it ever since.
15. A line requires a starting point and an ending point. Extend it infinitely in either direction and it becomes something that mathematics can describe but nature has never confirmed.
16. Giordano Bruno (1548 - 1600) said there was no time - only movement. Where there is no movement, there is no time. He was burned for ideas like this. The Church could not afford to let the model be questioned.
17. Time, understood dimensionally, is not a line. It is curved - rooted in the rotation of spheres, expressed in the cycle of light and dark, measured not by atomic vibration alone but by the living geometry of daylight.
18. A curved understanding of time does not contradict physics. It completes it.

IV. On Curvature

19. Before René Descartes (1596 - 1650), seafarers measured the world in nautical miles: degrees of arc along a curved equator, slices of a sphere. The curvature of the Earth was built into the measurement.
20. Descartes gave us coordinates: flat, orthogonal, elegant on paper. He made mathematics portable and powerful. He also made it flat.
21. The cost of that flatness has compounded over centuries. Space is curved. Light follows curved paths. The Earth is a sphere. Yet the coordinate systems we use to model all of these remain fundamentally straight.
22. Curved phenomena measured with straight instruments produce systematic distortions. Those distortions are then corrected with increasingly elaborate mathematics - rather than by returning to curved measurement itself.
23. ADI proposes that curved metrics, grounded in spherical geometry, offer a more physically honest foundation for measurement across all scales - from the sub-atomic to the astronomical.

V. On the Fragmentation of the Modern World

24. The same problem that afflicts physics at the cosmological scale afflicts engineering at the practical scale. Our infrastructure is fragmented - not because the systems are incompatible, but because the measurements are.
25. Airflow is measured one way. Heat another. Fibre another. Energy another. Cost another. Nothing aligns. Nothing scales. Nothing speaks the same language.
26. This is not an engineering failure. It is a measurement failure. Each domain inherited its own flat, linear metrics - and those metrics were never designed to speak to one another.
27. The consequence is waste: energy lost in translation between systems that could, in principle, be optimised together. The potential efficiency exists. The measurement framework to unlock it does not yet - or did not, until now.
28. A unified geometric framework - one that classifies systems by dimensional behaviour rather than domain convention - makes airflow, heat, fibre, energy, and cost comparable for the first time. It links physical geometry to economic performance across scales that were previously separate.
29. Dimensionality is not an abstraction in this context. It is infrastructure. It is the missing common language of the built world.

VI. On Curved Processes and Flat Tools

30. Every critical process in a modern data centre, and in the energy grid that powers it, is curved.
31. Air moves in vortices. Heat propagates spherically. Light bends, scatters, and refracts. Pressure fields expand in arcs. Even the demand curves of energy markets are curves, not lines.
32. Yet the industry measures all of these with flat SI units, as though curvature were a rounding error rather than the essential nature of the phenomenon.
33. Measuring a vortex with a linear unit is like measuring the circumference of the Earth with a ruler. The tool is not wrong. It is simply unfit for the shape of what it is measuring.
34. Curved metrics - units that encode curvature rather than eliminate it - allow these processes to be measured on their own terms. The result is not greater complexity. It is greater honesty, and with honesty comes optimisation that flat measurement cannot reach.
35. This is a practical claim, not merely a philosophical one. The difference between a flat approximation and a curved measurement, compounded across thousands of data points in a live system, is the difference between managed inefficiency and genuine performance.

VII. On Energy and Its Qualities

36. The energy crisis is, in part, a measurement crisis.
37. The industry treats energy as a single quantity - a number on a meter, a figure on an invoice. But energy is not one thing. It has qualities, and those qualities behave differently, lose differently, and optimise differently.
38. Electric energy is carried by electrons in motion. Thermal energy is the vibration of atoms and molecules. Mechanical energy lives in the movement of mass through space. Photonic energy is carried by light itself - by frequency, propagation, and the behaviour of photons at every scale.
39. These are not subdivisions of a single category. They are dimensionally distinct phenomena. Treating them as one number is precisely the kind of dimensional flattening that these theses are written to challenge.
40. When energy is understood as a dimensional system - four qualities, each with its own geometry, each optimised on its own terms - the economics of infrastructure change fundamentally. Losses that were invisible become visible. Efficiencies that were unreachable become accessible.
41. This is where physics and economics converge. Energy has always had qualities. We have simply lacked the measurement framework to act on them. That framework now exists.
VIII. On Light and Colour
42. Light is the most fundamental physical constant we know. It is also the one we understand most selectively.
43. The speed of light, as currently measured, is defined within a flat, two-dimensional electromagnetic framework. The spherical dimension - the dimension in which light actually propagates through space - remains outside the measurement.
44. Colour is what happens when light meets perception. It is the bridge between the physical and the experiential - and dimensionally, one of the most complex phenomena in nature.
45. A dimensional approach to colour and light opens the possibility of new instruments, new models, and new forms of understanding that current frameworks cannot accommodate. Colour is not decoration. In a dimensional system, it is DATA.

IX. On the Reformation Ahead

46. Every great scientific revolution has been, at its core, a reformation of measurement. Copernicus (1473 - 1543) did not discover new stars - he proposed a new centre. Newton (1642 - 1747) did not invent gravity - he gave it a language. Einstein (1879 - 1955) did not change the universe - he changed the frame.
47. The fragmentation of modern infrastructure, the inefficiency of modern energy systems, the unresolved tension between mathematical certainty and physical reality - these are not separate problems. They share a common root: the inherited assumption that flat, linear measurement is sufficient for a curved, dimensional world.
48. ADI proposes the next frame: one in which
• DIMENSION is taken seriously as a quality, not merely a quantity;
• CURVATURE is restored to measurement;
• ENERGY is understood as four distinct phenomena rather than one undifferentiated number;
• and the BUILT WORLD - its data centres, its materials, grids, its flows of heat and light and power - can finally be optimised as the geometric system it has always been.
49. This is not a rejection of what science has built. It is an invitation to examine the foundations - and to find, beneath the accumulated assumptions, something more honest, more complete, and more true to the shape of the world we actually inhabit.
50. Luther nailed 95 theses to a door. The number was never the point. The point was the nail.
These theses are offered in the spirit of inquiry, not authority. Like all reformations, this one begins not with answers - but with the courage to ask the questions that the prevailing orthodoxy has stopped asking.

Portfolio