Why Prompt Engineering Is Nonsense (opens in new tab)

(pavkam.dev)

14 pointssleepychu2y ago5 comments

5 comments

> “So why isn’t it engineering? Well, is asking a question “engineering”? Excuse me for being pedantic but from what I’ve learned it’s part of the field of “philosophy”. There is no engineering involved. And even if there is “skill” required to formulate a well-structured question for the LLM, that is not an engineering pursuit.”

Not a convincing argument when we’re talking about software.

Is designing a data structure engineering or philosophy? Go back 200 years and ask scholars whether determining the abstract properties of a non-physical object is engineering. They’d say “no” and point you towards Kant. Yet today that kind of thought process underlies engineered things that we call names like “infrastructure” and “products” even though they don’t fit the definition those words had as recently as 60 years ago.

eternalban2y ago

Data structures require mapping a conceptual model ("philosophy") to a representation by a physical machine. As a data structure engineer you will need to make engineering decisions such as alignment of data, data-precision (*), and behavior under different runtime regimes (single threaded, concurrent). All those are technical decisions that a mere 'philosophy' is not equipped to address.

* a hot topic in ML world actually. Precision required for the params.

jelkand2y ago

Philosophy is not a bucket where "soft" or "unskilled" knowledge work goes either. The author is doing themselves a serious disservice in not really explaining what about prompt engineering makes it not "engineering" and what about philosophy matches it better-- and I say this as someone who is on the caboose of the AI hype train.

cauliflower992y ago

Whether it is AI that overtake people's jobs, or people using AI that overtake people's jobs, I think the result will (and is) absolutely less available jobs.

It won't happen all at once (though the recent waves of layoffs might be a good argument that it is already happening). I think this is a slow burner that will phase jobs out over the long run:

- New and future openings won't be advertised where an AI tool can do the job instead. Think specifically in the digital art and text space. This means that people will find it difficult to leave their current job to find something better, leading to stunted careers and potentially much lower salaries.

- Current jobs will continue as they are (besides the current waves of layoffs) except for the consultant space. If ChatGPT can act as my personal Deloitte without the €€€'s and at a moment's notice, then of course I'm going to choose ChatGPT.

I see this already happening => https://old.reddit.com/r/blender/comments/121lhfq/i_lost_eve...

I agree with the article: "We don't know".

jschveibinz2y ago

The word “engineering” has been so diluted by its application to software job titles that the word on certain contexts has become ambiguous.

“Engineering is the use of scientific principles to design and build machines, structures, and other items, including bridges, tunnels, roads, vehicles, and buildings.”

This is Wikipedia’s definition, and I think it is pretty close. But I would add that “process” is a big part of engineering, as is attention to cost, schedule and resources. “Scientific” principles aren’t always required, but math and logic are frequently required.

Software engineering (forget about the job titles) is the well-defined process of using software resources (building blocks) to build and test cost-efficient software systems, on time and on budget.

If prompt language ever gets sophisticated enough to require the development of a system to construct it, then “prompt engineering” might be a thing.

But that would seem to be the opposite of the goal of AI: simplifying the user’s experience with arbitrary prompts (i.e. natural language)?

j / k navigate · click thread line to collapse