3D graphics from Resource Database – https://unsplash.com/@resourcedatabase

Generative texts: Working with Artificial Intelligence

How can you use artificial intelligence to generate creative and valuable texts? In this post, you'll learn more about the opportunities and challenges of generative texts.

Jairus Joer

Jairus Joer

The texts in this post were generated in parts by Microsoft’s Bing and corrected and revised by us.

Introduction

Generative content is already ubiquitous in our daily interactions online. We encounter it in the form of art, advertising, and a wide variety of content (including this post), and thus learn how we can make sense of programmatic-based generation of text for our own projects as well.

Tools

The selection of tools that produce generative texts is constantly growing. Depending on the scope of the project, the demands on the providers can vary in complexity, so that there is a suitable tool for most application purposes. Therefore, this article will not go into detail about the individual providers and their tools, but will focus on the general principles and methods of working with generative texts.

The following selection of popular tools is limited to five frequently mentioned providers and invites research, should your own project require it:

In this case, we chose Bing Compose to work with up-to-date data and sources.

Generate & modify

Prompts, or prompting, allows us to gradually adjust our requirements for the text and shapes the content generation parameters needed for generative texts.

For our demonstration, we generate a short description on this very topic, which should not be longer than 225 characters:

To preserve the generative nature of the texts, the texts were not translated in the following examples. A translation can be found in the section Translate

Prompt: Beschreibe generative Texte in maximal 225 Zeichen
Tone: Professional
Format: Paragraph
Length: Short

Preview: Generative Texte sind Texte, die von einem Computerprogramm erzeugt werden. Sie können verschiedene Zwecke haben, wie z.B. Informieren, Unterhalten oder Überzeugen. Generative Texte können aus Schlüsselwörtern, Daten oder anderen Quellen generiert werden.

By adjusting the tone via the interface, a modified version of the text is produced, which should communicate a certain enthusiasm for the topic to the reader:

Prompt: Beschreibe generative Texte in maximal 225 Zeichen
Tone: Enthusiastic
Format: Paragraph
Length: Short

Preview: Generative Texte sind Texte, die von einem Computerprogramm erzeugt werden. Sie können verschiedene Formen, Stile und Inhalte haben. Generative Texte sind kreativ, spannend und vielfältig. Sie können uns inspirieren, informieren oder unterhalten.

It is noteworthy here that the first sentence has not changed, only the continuing texts have been modified.

If the configuration features provided by the tool are not sufficient, we can specify our information via the prompt:

Prompt: Schreibe einen Teaser über generative Texte in maximal 225 Zeichen
Tone: Casual
Format: Paragraph
Length: Short

Preview: Generative Texte sind Texte, die von künstlicher Intelligenz erzeugt werden. Sie können spannend, informativ oder kreativ sein. Wie funktioniert das? Und was sind die Möglichkeiten und Grenzen? Erfahre mehr in diesem Artikel!

By modifying the prompt and changing Beschreibung to Teaser, the generated text changes to reflect a Call To Action for the reader.

Adjectives and other keywords can be used to further specify the modifications to be made to the text:

Prompt: 'Schreibe einen Teaser mit maximal 225 Zeichen für einen Blogpost mit dem Titel "Generative Texte: Arbeiten mit künstlicher Intelligenz"'
Tone: Informative
Format: Paragraph
Length: Short

Preview: Wie kann man mit künstlicher Intelligenz kreative und nützliche Texte erzeugen? In diesem Blogpost erfährst du mehr über die Möglichkeiten und Herausforderungen von generativen Texten.

Correct

The generated texts convey various impressions on the topic and provide a good basis for adapting them. It is already noticeable that some texts have exceeded the desired character limit of 225 characters. For further work with the text we focus on the last result from the section Generate & modify:

  • Origina ltext: Wie kann man mit künstlicher Intelligenz kreative und nützliche Texte erzeugen? In diesem Blogpost erfährst du mehr über die Möglichkeiten und Herausforderungen von generativen Texten.

  • Correction: Wie kann man mit künstlicher Intelligenz kreative und nützliche Texte generieren? In diesem Beitrag erfährst du mehr über die Möglichkeiten und Herausforderungen von generativen Texten.

Our corrections turn out to be very light, so we only adjusted two words to get closer to our own wording. Of course, the need for correction changes with the preceding parameters and the text length.

Translate

If we want to publish our content outside our language area, we can translate it into 30+ languages using DeepL.

The translation interface already offers precise customization of individual words or sentences, so that the desired result can already be generated and revised in the preview.

  • Original text: Wie kann man mit künstlicher Intelligenz kreative und nützliche Texte generieren? In diesem Beitrag erfährst du mehr über die Möglichkeiten und Herausforderungen von generativen Texten.

  • Translation (unmodified): How can you use artificial intelligence to generate creative and useful texts? In this article you will learn more about the possibilities and challenges of generative texts.

  • Translation (modified): How can you use artificial intelligence to generate creative and valuable texts? In this post, you’ll learn more about the opportunities and challenges of generative texts.

Use cases

As mentioned in the introduction, generative texts offer a wide range of use cases:

Generation

This post shows that generative texts can be used to generate and add your own content. In project terms, this can mean that text content can be generated in a shorter time with comparable quality.

Modification

In addition to obligatory A/B testing, existing (or generated) texts can be used particularly in inclusion. Tools such as SUMM AI allow reformulation of texts into simple and easy language.

Translation

Products and their data can be translated into a wide variety of languages with ease using DeepL, reducing manual translation work to proofreading in many cases.

In a more concrete example, we recently added English to this blog using DeepL to have large-scale translations done.

Ethics

Whenever generative content is used, this should also be communicated to the reader. The demand for transparency in this context is especially beneficial for text comprehension.

In the automation of work steps such as modification or translation, errors can always occur which would normally be eliminated in a manual process.

The decision on transparency lies with the content providers and can in its simplest form already look as follows:

The texts in this post were generated in parts by Microsoft’s Bing and corrected and revised by us.

TL;DR

artificial intelligence can be used to generate, modify, and translate texts. In addition to use cases, the ethics of generative texts are discussed, among other topics.