I have to admit, this post is a little ironic given that I am literally charging an organization for a workshop on prompt engineering tomorrow, but the truth is it is a little weird to charge anyone to teach them how to prompt engineer.
Teaching someone to prompt engineer is a little analogous to hiring a personal assistant then taking a class to ask them to make you a coffee.
Sure, that is a little unfair. Maybe this: prompt engineering is more like hiring a web designer to teach you to build a website, then needing a class to talk to them.
Still unfair, I agree. Let’s say this time you hire a translator to help you learn and communicate in a foreign language. Again, do you really need a class to talk to your translator?
My guess is those involved in teaching and training with AI have a sense of their nakedness, a sense that the emperor, or in this case the AI workshop facilitator, has no clothes. In practice, for experienced users of AI, the AI is probably better at teaching users to use it than the workshop facilitator.
This sense of nakedness may have been what has created this proliferation of useless acronyms online to teach people how to prompt engineer, a sort of overcompensation or effort to carve a niche where none is necessary. Here are just a few of the countless available:
- RACE: Role, Action, Context, Execute
- CREATE: Character, Request, Examples, Adjust, Type, Extras (Seriously WTF)
- FACTS: Formulate, Acquire, Create, Type, Scrutinize (Why is Type a thing in two of these lists)
- CARE: Context, Ask, Rules, Examples
- RISEN: Role, Instructions, Steps, End Goal, Narrowing
- SSSSS: Set the Scene, Be Specific, Simplify the Language, Structure the Output, Shape Feedback (sssssSeriously?)
- Relic: Role, Emphasis, Limitation, Information, Challenge
Check it out for yourself – these go on and on.
But there is one that is so special, that is so incredibly overdone and the opposite of all pragmatism, utility, and decency that it deserves to be displayed here in all its glorious fullness. Behold:
Please stop looking at that. How pointless. Seriously, BAB?
Here is the deal; AI is a weird, jagged technology. It can be amazingly helpful but then fail unexplainedly; it hallucinates; it will tell you your “French fries into salad small business idea” is genius (thank you South Park); it can also be inconsistent. It has tendencies and a “personality” that, once you get to know it, allows you to get more out of it. Its quirky and it helps to know the quirks, but all these prompt engineering techniques are not the way.
Here is a better way. Meta Word Salad. (Not to be confused with Laker’s legend Metta World Peace, formerly Ron Artest.)
Here is my whole prompt engineering workshop: Give it word salad for whatever you need, then ask AI to engineer the prompt (the Meta part). Fix it in post. That’s it. Meta Word Salad.
Even simpler, just think of this as vibe coding your prompts.
In Co-Intelligence, Ethan Mollick writes about 3 sleepless nights, about what happens when a user realizes what AI might be capable of and spends around 36 hours with it. This is not Gladwell’s 10,000 hours: if it were maybe some of these ridiculous acronyms would be useful. We are talking 36 hours, just to get a sense of what it can do and what it can’t. To internalize some of its quirks, to get a sense of how to get what we need from it. Sure there are some tricks, but it is just not that complicated.
At AuthenticALliteracy we have some simple focuses. We want new AI users to want to use AI. I can’t imagine anything more discouraging to the spirit of wonder using AI can evoke than trying to ploddingly write a prompt using one of these acronyms. Moreover (I am an English professor and can’t help but write this way – not AI generated), using these lists gets in the way of helping users actually do the work to understand what the AI is and is not capable of. They truly are the opposite of what any good AI training would try to spark in their audience. We need wonder and exploration, not this tedium.
We also want AI users to get faster, better results than they would without it. We are trying to be more efficient, which means we don’t have time to write a beautiful prompt. Just spill, tell it with errors and malapropisms what you want, and do it quickly. Just get it out. It will fix the prompt for you. I cannot state what an unimaginable waste of time it would be to memorize ANY of these acronyms and then to try to write prompts using them.
If you understand the tool, you might have an insight for when giving AI a role might be helpful, or when context is important and when it isn’t, or when it is critical to specify the exact outcome you want. Using these kinds of prompts is a barrier to developing these authentic insights because it doesn’t give you an intuitive sense of the LLM’s tendencies.
Meta word salad. Just go play. Give the LLM your vibes and make it write the prompt for you. If it doesn’t do what you want, tell it it is bad and to do better. Or ask it what you need to do to get what you want. This is how to build your AI literacy and competence and is the quickest way to building prompts you are actually likely to use.
Leave a Reply