The Missing Guide to Prompt Engineering (opens in new tab)

This looks AI generated.

There are places where it includes citation references like [15] (search for "on complex problems" on https://appetals.com/promptguide/advanced-techniques-for-pow... ) that are just text, they don't resolve to anything.

Other places repeat the same point twice for no apparent reason - the "Context window" section on https://appetals.com/promptguide/understanding-the-machine-m...

I spotted a few "not X but Y" constructions, like "You'll learn to communicate with AI like a pro, not like someone shouting into the void."

I don't think reading this is a good use of your time.

I'm a bit tired of prompting guides that basically revamp https://arxiv.org/abs/2406.06608

There's other criticism here about the writing quality (AI generated) but it's also quite out of date content-wise.

> Start with a basic prompt >Add Chain-of-Thought reasoning >Implement self-criticism >Add format constraints >Compare the quality and reliability of results

The 'techniques' are incredibly basic and Chain of thought isn't really relevant today considering there are models with COT 'baked in' i.e. the 'thinking mode' class of models.

Why have so many non-engineering things become "engineering"?

Prompting -> "prompt engineering".

"sudo make me a sandwich" -> privilege-escalation engineering while process engineering a food engineer.

Prompt engineering isn't going anywhere. People keep saying it's dead, that the next model will make it obsolete. They said the same thing about ChatGPT 4, then Claude, then Gemini Pro. Even though the AI models are becoming more powerful each day, a poorly prompted AI response could hinder your ability to get the right results.

> This is prompt engineering. Though the term suggests something mechanical, something reducible to formulas and best practices, the reality proves more nuanced—perhaps more human—than we'd care to admit

Ah nothing like a double emdash early to know that the page is not worth reading.