A little bit more time and people might finally understand why we have formal deterministic languages, and we'll be back full circle to proper programming.
What jumps out at me is a lot of this is still very task oriented. And each to their own, but anecdotally, I haven't seen great results from task oriented behavior.
I don't mean that it does not produce what was asked for. I'm saying that tasks even when created by engineering and product teams are often wrong.
I lean very heavily towards outcome based prompting. Say exactly what do you want achieved and then maybe give some constraints, ie. what definitely not to do.
In my experiments, this has always produced much, much better results.
Interestingly, it's less engineering and more customer focus.
> I lean very heavily towards outcome based prompting. Say exactly what do you want achieved and then maybe give some constraints, ie. what definitely not to do.
I do this when writing stories/projects/issues/epics for humans. Works great.
If you read any management book published in the last ~70 or so years, you’ll find that “Make sure people understand the goal” is the ultimate hack. They even use this in militaries!
“go take that hill” works a lot better than “walk 50ft to the right and shoot at those bushes”. You always get what you ask for :)
Say what you want, followed by how you want it done, which includes unit testing, research (Claude can do research for you) on best practices, etc. I usually also follow up with "Why would I NOT want to do this this way, be critical of this." Usually outlines sensible issues, you give guidance, then the plan is ready to go. Then you test, test, and when you're done testing, you test some more, and review all code. I find that despite how busy that all sounds, I save myself weeks of effort.
> “I want to know why my React app’s state is not updating when I click a button.”
After:
> “React 18. useState. Button click handler sets state but component does not re-render. No error in console. Explain top 3 causes and fix for each. Show code.”
> Notice the transformation: 22 words down from a long conversational sentence, yet more information is packed in because every word carries signal.
It's 27 words up from 17, and would produce poor results on the local models this claims to be targeting. Without some way to iterate and close the loop, models are pretty bad at producing good prompts.
It still makes me laugh when i see "prompt engineering". I open articles posted on here that contain many diagrams and novel jargon, all for it to amount to using a fucking markdown file with some text in it.
A little bit more time and people might finally understand why we have formal deterministic languages, and we'll be back full circle to proper programming.
Exactly. If LLM coding had come first we'd build statues of whoever invented deterministic programming.
LMAO
LLMAO, surely.
This is counterproductive, you can't reduce eloquence to a list of examples. Read good literature is still the best advise we can give here.
Prompt engineering keeps rediscovering "say what you mean" with extra steps.
Snake oil. Tell me something that has hard irrefutable and reproducible evidence.
Very interesting read.
What jumps out at me is a lot of this is still very task oriented. And each to their own, but anecdotally, I haven't seen great results from task oriented behavior.
I don't mean that it does not produce what was asked for. I'm saying that tasks even when created by engineering and product teams are often wrong.
I lean very heavily towards outcome based prompting. Say exactly what do you want achieved and then maybe give some constraints, ie. what definitely not to do.
In my experiments, this has always produced much, much better results.
Interestingly, it's less engineering and more customer focus.
> I lean very heavily towards outcome based prompting. Say exactly what do you want achieved and then maybe give some constraints, ie. what definitely not to do.
I do this when writing stories/projects/issues/epics for humans. Works great.
If you read any management book published in the last ~70 or so years, you’ll find that “Make sure people understand the goal” is the ultimate hack. They even use this in militaries!
“go take that hill” works a lot better than “walk 50ft to the right and shoot at those bushes”. You always get what you ask for :)
Say what you want, followed by how you want it done, which includes unit testing, research (Claude can do research for you) on best practices, etc. I usually also follow up with "Why would I NOT want to do this this way, be critical of this." Usually outlines sensible issues, you give guidance, then the plan is ready to go. Then you test, test, and when you're done testing, you test some more, and review all code. I find that despite how busy that all sounds, I save myself weeks of effort.
Prompt engineering: talking to autocomplete like it owes you money.
Ironic. This article violates all rules that it intends to set out.
It is easy to spot AI generated garbage.
"Premium models: GPT-4o, Claude Sonnet, Gemini 1.5 Pro"
TFA was written by an AI without even search access, or it would know that all those are OLD deprecated models. AI Slop.
If the author doesn't even bother checking what his AI spit out, why would I read it? Useless article. I'm baffled that this reached the front page.
Not to mention the very first example.
Before:
> “I want to know why my React app’s state is not updating when I click a button.”
After:
> “React 18. useState. Button click handler sets state but component does not re-render. No error in console. Explain top 3 causes and fix for each. Show code.”
> Notice the transformation: 22 words down from a long conversational sentence, yet more information is packed in because every word carries signal.
It's 27 words up from 17, and would produce poor results on the local models this claims to be targeting. Without some way to iterate and close the loop, models are pretty bad at producing good prompts.
It still makes me laugh when i see "prompt engineering". I open articles posted on here that contain many diagrams and novel jargon, all for it to amount to using a fucking markdown file with some text in it.
So, me think. Why waste time say lot word when few word do trick?
Well a model is not your buddy, there is no need to chat with it.