Why does the quality of your AI prompts determine the quality of the output?

Quick answer: an LLM is optimised to produce grammatically plausible text, not factually correct text — it doesn't inherently know the difference between "the dog walked down the street" and "the dog flew down the street," both are valid English. A vague or under-specified prompt gives the model more room to fill gaps with something plausible-sounding but wrong, so prompt quality directly controls how much of that gap-filling happens.

Prompt engineering is trying to get the right answer back out again, and that can take a lot of work on our part to get the correct answer without any hallucinations, or "boasting" as we might call it if it were a human. That's a slightly different thing to machine learning — machine learning is really quite niche and locked down to a specific task, but this does mean that LLMs have quite strange limitations.

They can create plausible stuff — after all, an LLM is originally designed as a translator, and its whole purpose is to create a line of text that's correct from a grammar point of view, not correct from a context point of view. You can say "the dog has walked down the street" — that's complete, true, and plausible. "The dog is flying down the street" is also valid as English, but complete rubbish, because we know that dogs can't fly. The computer doesn't necessarily know that, so it can make stuff up — it doesn't really have true awareness. There's a lot of things we infer just by being alive that an LLM doesn't know about, because it's only ever read stuff.

They're also very computationally expensive to run, because they're doing lots and lots of calculations — we've modelled the brain, and brains are very efficient with their energy, whereas computer silicon isn't. And, like I've already mentioned, LLMs require careful prompt inputs in order to get the right response out.