Shaping AI Conversations with Prompt Engineering

Few things are as deceptively simple yet profoundly impactful as a well-crafted prompt. Following last week’s article 10 ChatGPT prompts every customer success professional needs, a few readers responded with the same question: how do you know what kind of prompt to use in every situation?

Enter “prompt engineers,” a seemingly faddish job role that lies at the intersection of language, logic, and machine learning and can garner 6-figure salaries at top tech companies. No, the job description isn’t just “have great conversations with ChatGPT all day.”

A prompt engineer is part psychologist, part linguist, and part technologist. Their role is to understand both human intent and machine logic, and then craft a prompt that seamlessly bridges the two for the right application. This means they need a deep understanding of user goals, contextual subtleties, effective outputs, and the underlying AI's capabilities and limitations.

The Multifaceted World of Prompts

Prompts aren’t just questions; they're designed instructions. Let’s look at some of the factors that are important when writing a good prompt:

• Open-Ended vs. Close-Ended: Just like a survey might prompt a human with “answer true or false,” LLM prompts can ask for a very specific response that leads to the next step in a conversation chain.

• Adding dimensions: Prompting with “tell me a story” will provide a much different response than “tell me a story set in Victorian era England about a detective that solves cases with his quick wit and attention to detail.” The more context you provide about tone, tenor, scene, personae, etc. the more specific the response becomes.

• Stipulating Format and Length: Dictating the structure or length of a response can be vital. "Describe the solar system" could result in a varied-length discourse. But "In three sentences, summarize the solar system's main features" ensures brevity and conciseness.

• Active vs. Passive Tone: The tone you set in the prompt can affect the assertiveness of the response. "Can you tell me about photosynthesis?" may yield a more reserved answer compared to "Explain photosynthesis in detail."

But Why Does It Matter?

Last week, OpenAI dropped a huge stunner that ChatGPT can now see, hear, and speak and users have tested everything from describing IKEA instructions to translating ancient Catalonian texts. As AI continues to permeate our lives, the art of crafting the perfect prompt will only grow in significance. And yes, the job listings are reflecting the diverse challenges.

Check out this job posting by Anthropic, which is an open-ended prompt engineering job that admits “Given that the field of prompt-engineering is arguably less than 2 years old, this position is a bit hard to hire for! As a result, we ask that you share with us a specific prompt engineering project on LLMs that you’re proud of in your application!” The proposed pay is $250-375k a year.

And again, it all comes down to what you’re trying to achieve with your prompts. If you’re a developer trying to work LLMs into your code flow, the prompts will be far different than a sales leader trying to generate customer emails. Imagine coding something like the below pseudocode:

feedback = user.response;
if (prompt(feedback, "Evaluate the sentiment of this response to our latest release, returning one of: positive, negative") == "positive") then:
  celebrate();
else:
  panic();

Suddenly we’ve unlocked a whole new world of code statements that not only evaluate and analyze, but do so right inside the core decision functions with language (and image) understanding.

As Anthropic said themselves, the field of prompt-engineering is nascent but exploding with possibilities. The AI landscape necessitates expertise in shaping precise prompts that extract value from sophisticated language models. Whether you're a tech enthusiast, a linguist, or someone just fascinated by the crossroads of human and machine communication, the science of prompt engineering heralds a transformative phase in how we integrate AI into diverse sectors and applications. The real power isn't just in the AI itself, but in the dialogue we design with it.