THE PROMPTS FOR ARTIFICIAL INTELLIGENCE
The consolidation of artificial intelligence in everyday life is not only a technological leap, but also ushers in a new era of convenience, efficiency and automation that changes the way you interact with the world around us.
But in order to go deep and use artificial intelligence in the most effective way possible, the commands (prompts) we set are especially important to have the best result. They are the language through which humans give their instructions to machines.
Examples of basic commands might be, “Explain how photosynthesis works” or “Write a summary of Hamlet,” but these prompts only scratch the surface. The generative language model can do much more, and the knowledge of writing good commands helps with that. AI can produce complex answers based on the millions of texts that the user has entered. Another way to use it is to create applications or to automate a large part of the work day. This is achieved with effective commands.
Ben Stokes, the creator of PromptBase, a marketplace for AI commands, says:
“A good command writer can create commands that produce consistent, high-quality output (e.g. images, text, or code) at low cost.”
But what makes a command good? Look at the image above. Think about how you could describe it. “Smiling man in suit giving a thumbs up”? “Man with white teeth, thumbs up, comic, speech bubble”? A good start, but describe the image in more detail. Consider keywords such as “Caucasian” or “handsome.” You can define the style as “vintage” and “pop art”. All of these words are already contained in the command for that particular image.
To significantly improve the image generated by the DALL-E 2 model, researchers are working on refining the instructions and developing “supercommands.” These complex instructions, consisting of hundreds of words, prompt the artificial intelligence to explore its database and expand its capabilities beyond the imagination of the creators. One approach involves role-playing with ChatGPT, where users prompt the AI to pretend to be someone else. By using specific words and sentence structures, the AI’s behavior can be influenced. More complex and detailed instructions lead to better results. In return, ChatGPT may ask for more details.
Like a journalist using different words to ask the same question, words and sentence structures can influence AI behavior. How you phrase the instructions is important, not just the content. More complex and specific instructions lead to better results.
Last but not least, ChatGPT can become a command writer. But why learn to write good commands if the AI can do it on its own? Because while common prophecies suggest that AI will take control over humans, most AI applications will long rely heavily on the “human factor” to provide instructions, refine results and guide technology in new directions.
The more we use artificial intelligence, the more we understand what works, what doesn’t, and why. This also makes it less threatening—ChatGPT is still a limited machine, but educational institutions should jump at the chance to get involved in learning to write AI commands while the field is still in its infancy.
Digital skills for all