Prompt engineering is not a thing. If two people type the same prompt and get different code, the prompt was ambiguous. That is not programming. With Shree Manickaraja and Cam on Compile After Class. [

I think it's actually kind of an enormous problem that cloud AI platforms insert fake entropy so it looks like it's not deterministic and so partly thwart really understanding the chain of causality in minor prompt tweaks.

Yes waiting on you guys ie microsoft to invent a deterministic english programming language were certain statements can only mean one thing The English language is far to amnigous as to many terms mean more than one thing. That is the next logical step for me

I was reading an agent skill source file and thought it would be nice if it was code instead of natural language.

i can use prompt engineering to make a model say the same thing, dont know about the same code

If the output is the same but the code different, does it really matter?

if the desidered outcome is correct and quality decent than it's irrelevant. it cannot be exactly the same, since an LLM is not a deterministic system

Any system with human involvement will bring ambiguity..LLMs must not be exposed to humans directly...a for loop is compiled and then executed, compiler throws an error if u wrote it ambiguously..similarly Human inputs must be validated deterministically before sending it to LLM

If you write a for loop in C, it does not mean that it will run in the same way unless it uses the same compiler. You can write a compiler that will compile a for loop into a different set of instructions.

Classic consultant answer... It depends. 1. Temperature... Set it to 0. More predictability. 2. Only when model stay the same. Retune of changed and test. 3. Same hardware? 4. Specificity wins over generality. Narrow down what you want. More predictable but not 100% deterministic

The difficult concept here is “prompt engineering” implies a logical input. You’re used to a prompt parsing out method calls, ls grep pwd SELECT INSERT and such. Engineering implies logical flow of data too. Prompt engineering is simply rubber duck programming where the rubber duck responds and codes the thing. So, it’s conversational and replies back with logic that you refine (hopefully) and come to a solution together.

dual wielding ambiguity and tech debt is valid as fuck! we twice around insane ! 0o0/b

Promtgramming?

What if one person uses the same prompt and gets different code each time?

too much spice love your content

That's because unlike with local models, with cloud models you don't get control over the seed and a bunch of other knobs to tweak an turn. If the prompt is the same and the settings including the seed is the same, the output is the same. Fixing the seed and changing the prompt is one of the easiest ways to see if a thing in the prompt changes the result for better or for worse or if at all.

Fake news

Do don’t even talk to me about coding. It’s over, doesn’t matter your logic.

It’s dsnt matter . If it works it works. Who care if the code is different wald by the way you can produce the same product by asking a specific code langage so yes prompt engineering it’s a thing very real. And the faster you realized That is is the better

If two software engineers solve the same problem with slightly different code, is that not software engineering?