The AI Elixir of Dr. Doxey: Myth #2 – “You don’t need prompt engineering anymore, you just talk to GPT.”

By Sébastien Wiertz, November 17, 2024

Welcome back to the Wild West of AI myths and misconceptions! In my last article, I channeled my inner Lucky Luke to debunk the myth that ChatGPT and similar tools can resolve mathematical problems like a calculator. Spoiler alert: it can’t. Instead, it predicts responses based on patterns it has learned. If you missed that one, I highly recommend reading it first ( here is the link) , it’s the foundation for today’s myth busting session.

This time, let’s tackle a statement I’ve witnessed : “You don’t need prompt engineering anymore, you just have to talk to GPT ! ”
My answer could have been “Wow, that’s amazing, thanks for this information!” but knowing how the system works, I will try to lasso this runaway notion/nonsense. AI may be powerful, but it’s not magic. Just like in the first myth, there’s a grain of truth here, but the reality is far more nuanced and far less miraculous.

Roping in the Truth: A Demonstration Rodeo

So, you can just talk to GPT? Sure, you can. And for simple tasks like “What’s the definition of Semantic web ? or “Give me a fun fact about illusionists”, you will experience that it works well (even if sometimes you will find out that it is better to write down your prompt for clarity).

So what’s your point Sebastien ? It works, doesn’t it ? yes … if you want simplicity and stay at the surface of the true power but the minute your query becomes more complex or requires nuance, context, or specific instructions, you are riding into trouble without a well crafted prompt. Why is it so (still) ? Because prompting is not just about asking, it’s about asking effectively. I explained the same story 20 years ago with google searches. It is not different today. It is just a different tool.

In my previous article, I explained how neural networks process tokens, breaking down your input into chunks, converting those chunks into vectors, and finding patterns to generate a response. The meaning of each token depends on its neighbors (semantic understanding), and therefore context is the sheriff in town. When your input lacks context, the AI flounders like a herd of cows running wild, scattering in all directions with no clear path. Let me give you a concrete example to explain how the concept is important in my prompting.

Think of the neural network as an eager student in class. If you give it a vague question, it’ll guess what you want, and that guess might be way off the mark. But if you guide it with clarity and intent, it can transform into the sharpest shooter in town.

Here is an example:

Casual Prompt: “Explain quantum entanglement.”

Response: The neural network might say something like, “Quantum entanglement is when two particles are linked, so what happens to one affects the other, even if they’re far apart.” It’s correct but surface level, something you could copy/paste from Wikipedia. (so no need to burn expensive energy on GPU to get such a result #staygreen)

Let’s now refine the prompt.

Engineered Prompt: “I’m an engineer with a solid understanding of quantum mechanics. Explain quantum entanglement, emphasizing its role in quantum computing and referencing Bell’s Theorem.”

-> Now we’re diving into deeper waters: GPT will deliver high-level information tailored to your expertise, skipping the oversimplifications and speaking to you as an engineer.

Alternative Prompt: “I’m a 14-year-old trying to understand quantum entanglement. Could you explain it like you’re a friendly teacher?”

-> Here, GPT simplifies the explanation and adjusts its tone, adopting the approachable style of a “friendly” teacher.

See the difference? The first prompt gives you a vague, generic response. The second and third prompts guide the AI to deliver answers tailored to the user’s expertise, purpose, and tone. This is what prompt engineering is all about: not just asking questions but steering the AI to give you exactly the depth, focus, and clarity you need. It’s the difference between riding a wild horse and guiding a trained one, you have the same horsepower, but one gets you where you need to go 😊.

So, why do some people believe you don’t need prompt engineering anymore? Well, because it’s partially true for casual or everyday queries. AI systems are designed to handle natural language inputs, so it feels like you’re just having a conversation. But this simplicity is deceptive. For more complex tasks whether it’s analyzing texts or summarizing complex contents for a specific audience, prompt engineering is the secret sauce that separates amateurs from pros.

Digital illusionists love to downplay this because it makes AI sound more approachable. “Just talk to it!” they say. But here’s the thing: when you talk to AI without clear instructions, you’re playing roulette with your results.

Let me rephrase it (again): AI is not some mystical oracle that understands your intent perfectly no matter how you phrase it. It’s a tool, a powerful one, yes, but only if you use it right! Imagine handing a lasso to someone who’s never used one and saying, “Catch that cow!” Without knowing how to swing the rope, aim, or time the throw, they’ll end up tangled in their own lasso while the cow wanders off into the sunset (🐄 🤠)

Prompt engineering isn’t a relic of the past, it’s a critical skill for getting the most out of AI. Sure, for basic questions, you might get away with casual phrasing. But for anything meaningful or nuanced, prompt engineering is your best friend. It gives you control, precision, and clarity. Qualities that make the difference between you and everyone else relying on vague inputs. So spend some time learning to prompt properly, craft carefully the context and your level of knowledge in order to make the difference. Yes, you have to use your brain to gain the best insights from the AI “brain”. The moment it changes, it would mean that AI would think like us, that day, I will start worrying but it will not be happening anytime soon (I keep this other myth for a next article 🤠)

So, the next time someone tells you, “You don’t need prompt engineering anymore” just smile, you know you are getting more juice out of the tool than that person does.

In this Wild West of AI, it isn’t about who shoots first. It’s about who shoots with precision and mastering prompt engineering will make you the sharpest shooter in the neural network saloon.