Good Prompting tips vs bad tips

Your students will have to learn the difference between sound advice and fragile hacks.

You’ll find plenty of advice online about how to prompt an AI, but not all of it is worth following.

Some of this advice even comes with studies to back it up.

The catch? Measuring the effectiveness of prompts is incredibly difficult. These studies often work well for tasks with a clear "right" answer, like solving math problems or debugging code. But for more open-ended tasks—like "design my LinkedIn content strategy for Q1" or "give me five ideas to boost awareness for my startup"—things get a lot trickier.

Tasks like these require nuanced judgment and input from multiple human experts to evaluate properly. This makes experimentation more expensive and complex, which is why most studies don’t even attempt to tackle them.

The Problem with Hacks

Many studies also frame their comparisons poorly. Hacks are often:

Tested against basic prompts, usually using one-shot or few-shot techniques.
Rarely compared to well-optimized prompts that respect the 3 criteria of good prompting.

Principles vs. Hacks

This brings us to an important distinction: Principles vs. Hacks.

Hacks rely on exploiting quirks in large language models, like:

Tokenization tricks.
Specific phrasing patterns.
Edge-case behaviors in the model’s training data.
They can deliver small, specific improvements but are often fragile. A model update can render them useless.

Principles are based on how large language models are fundamentally designed and trained:

They are sustainable, adaptable, and effective across a wide range of tasks.
They remain reliable even as models evolve.

For example, a study might claim that a specific hack, like "tipping," works. But its impact will likely be minor compared to applying strong, fundamental prompting principles (the ones you’ll find throughout this guide).

The Takeaway

Hacks can sound clever, but they rarely deliver meaningful results if you’re already following good practices. True mastery comes from:

Understanding how the AI works at its core.
Applying principles that stand the test of time.

Focus on the fundamentals! They’ll serve you better in the long run.

LLMs: What is an LLM?

As a trainer, you need to clearly explain what an LLM is at its core—and just as importantly, what it is not!

3 Criteria of a Good Prompt

As a trainer, teach the three key criteria for assessing prompts: focused, guiding, and dense.