Prompting as an interface between humans and machines

In the fast-paced world of artificial intelligence (AI) and machine learning (ML), ‘prompting’ has come to play a key role. It defines how people can interact with machines and make effective use of their capabilities. For software developers, it is particularly exciting to observe the development of this technology and identify future trends. This article highlights current developments and offers a glimpse into the future of prompting.

End users as prompt engineers: Chat mode is currently the most widely used method for interacting with AI. A prompt is essentially a specific question or instruction, which can be supplemented by frameworks and leads to communication with the AI. Precise prompting plays a crucial role here, as different word choices can lead to varying results. One of the challenges is the reusability of prompts and the transparency of AI decisions. For users with less technical expertise , there are libraries and plugins designed to simplify these processes, but these are usually only used by a limited target audience.

Challenges for end users when prompting:

• Language barriers: English is often the language used for training, whilst other languages – particularly those that are less widely spoken or more complex in structure – often receive less training data. This means that queries in these languages are either misinterpreted or answered less accurately.
  
  
• Effectiveness of prompt formulation: The structure and wording of a prompt have a significant impact on the quality of the response and often need to be adjusted. If the context is missing or key terms are not clearly defined, there is a risk that the model will provide irrelevant information or omit important details.
  
  
• Complexity of iterations: Without a technical understanding of parameters and how to repeat the process effectively, it can be difficult to create successful prompts. If end users do not understand how to modify their queries or which parameters they need to adjust to obtain relevant results, the process becomes tedious and frustrating. In many cases, iterative adjustment is required, which increases the complexity.
  
• Lack of transparency: Often, the results, the underlying sources and data protection for end users are inadequate. It is not always clear to end users which sources have been used or how securely their data is being processed.
   

In many cases, prompting already takes place seamlessly in the background, so that users do not even notice the actual inputs. Instead, they have access to simple parameters that are embedded within existing programmes. This offers greater stability and optimisation for specific use cases by reducing unnecessary complexity. This is how generative artificial intelligence features in our everyday lives without us even realising it.

Examples:

• Android – Photo editing: Users can edit photos using simple commands such as ‘Adjust brightness’ without having to adjust complex settings.
• Android – On-screen translation: With a single click, the text on the screen is translated into another language.
• iOS – Creating reminders: Voice-controlled reminders, e.g. “Remind me to take out the rubbish at 9 am tomorrow”, make the experience particularly intuitive.