In our digital world today, a new type of influencer has risen to prominence — the "prompt influencer". Promising shortcuts to effectively command AI models like GPT, they have become a regular sight on platforms like Twitter. However, the secret they often sell is quite simple: there are no secret prompts. Effective prompting isn't a product of mysterious magic, but rather, a consequence of understanding and practice.
Misconceptions About Prompting AI
Often, people perceive prompting as the key to unlocking the full potential of AI. But there are reasons why we should look beyond this:
Transient Relevance of Prompts: AI systems are increasingly capable of intuiting user intent. The importance of specific prompts is diminishing with AI models like GPT-4 and Bing becoming more sophisticated in understanding user input.
Overemphasis on Superlatives: It's common to see prompts using superlatives, such as "Act as the smartest person ever", with the assumption that these grandiose directives result in superior AI performance. However, empirical tests reveal no significant improvement in the output quality.
The Interactive Approach: Instead of relying on perfectly crafted prompts, engage with the AI in an iterative manner. Ask for something, review the output, and then ask for adjustments. The more you interact with the AI, the better the outcome.
Practical Tips for Prompting
Despite the myths, there are certain techniques that can make your experience with AI more productive:
Context and Constraints
Large Language Models (LLMs) operate based on prediction; they anticipate the next word that follows your input, akin to a sophisticated autocomplete function. To achieve more engaging results, disrupt this pattern by providing context and constraints.
Assigning roles to the AI, such as a teacher or a circus clown, can tailor the output to suit your purpose. You can also add stylistic constraints like "write this in the style of the New Yorker" or "avoid repetition". Providing the AI with existing information or data you have can also be beneficial.
Programming in Prose
For advanced prompts, think of the process as programming in plain English. You give the AI a series of logical, step-by-step instructions. Though it may not always be consistent due to inherent randomness in AI outputs, it's a worthy approach.
One effective technique is the 'Chain of Thought' prompting, which instructs the AI on how you want it to reason before you make your request. This method can improve the accuracy of your prompts.
Using Magic Words
Certain phrases do tend to yield better results by modifying the context of the answer. Here are a few to experiment with:
"Be creative/make any assumptions you need."
"Show your work/provide sources/go step-by-step."
"Write me code and tell me how to use it."
"Write a draft/provide an example."
These can help nudge the AI to generate novel outputs, provide more transparent and accurate responses, create functional code, or deliver draft content for your projects.
The Power of Practice
The most effective way to master the art of prompting is simple: practice. Engage with the AI, use it frequently, experiment with different types of prompts and engage in a back-and-forth dialogue. Be persistent, patient, and open to learning from the outcomes. Over time, you'll build an understanding of the AI's capabilities, limitations, and the types of prompts it responds to best. Remember, as much as you are learning, so is the AI. Stay updated with the latest AI developments and you might uncover new ways to tap into its potential.