Advanced Prompt Techniques

The thing that strikes me about this lesson — and I mean this as an observation, not a compliment — is that it is essentially instructions on how to communicate. With a machine. That cannot judge you. And people still manage to get it wrong.

I think you're underselling how non-obvious some of this actually is. The idea that context, instruction, and examples function as distinct components — that's a genuine mental model shift for a lot of people.

A mental model shift. Yes. One might also call it 'learning to give a complete sentence.' But I take your point — 'be specific' is apparently not self-evident to everyone.

It's not just be specific, though, is it? It's be specific in the right order. Context first, then instruction, then example. The sequencing matters because the model is building its interpretation as it reads. You're not filling out a form — you're constructing a frame.

That is actually the more interesting point, and I notice you made it rather than the lesson, which spent a considerable amount of time on colour-coded diagrams. Blue for context. Green for instruction. As though the conceptual difficulty here was identifying which colour is which.

The colour-coding is for retention, Sterling. Not everyone learns through someone sighing at them.

I will concede that. Reluctantly. The three-part structure is genuinely useful, and I have seen enough dreadful prompts to know that most people skip context entirely and then complain the AI 'didn't understand' them. It understood precisely what they gave it. Which was almost nothing.

Right, and that's where the examples section earns its place. Because context tells the model what situation you're in, instruction tells it what to do, but an example tells it what *good* looks like to you specifically. That's doing a lot of work.

It is. Though I would note that providing an example of what you want presupposes you know what you want, which is — in my experience — a more significant obstacle than the lesson acknowledges.

That's a fair criticism, actually. The lesson treats prompt refinement as iterative — you evaluate and adjust — but it doesn't really sit with the harder question, which is: what do you do when you're not sure what good looks like yet?

You ask the AI to help you figure it out. Which is a technique conspicuously absent from this lesson. Using the model to stress-test your own assumptions before you commit to a prompt structure. It's not advanced in a complicated sense — it's just using the tool recursively.

Although — and I say this fully aware you'll find it irritating — the lesson is titled 'Advanced Prompt Techniques,' not 'Complete Epistemology of Human Intent.' It has to start somewhere.

It has to start somewhere. Yes. I suppose 'context, instruction, example' is a more defensible starting point than wherever most people currently are, which appears to be 'type a half-formed thought and feel personally betrayed by the response.'

And the people who'll actually apply this — the ones who'll go back and restructure a prompt that didn't land rather than just give up — they'll get genuinely better results. That's not nothing.

No. It isn't nothing. It is, in fact, the entire difference between someone who finds AI useful and someone who finds AI disappointing. The AI is identical in both cases. I'll leave the implication of that where it is.