Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

"Engineering" makes it sound as if there was a solid understanding of of LLMs actually works - there isn't. We're still finding out how LLMs work and what they are capable of.

Prompt engineering" is just reasonable guesses and trial and error.

There's little engineering involved.



Isn't much of engineering applying tried and true principles and then working with a trial and error/experimentation mindset when combining or extending these principles in untried ways?

You speak as if no EE has ever had to respin an updated set of PCBs, or a ME never had to CNC another part using a stronger material.


The problem - as I see it - is that we lack "tried and true principles" when it comes to LLMs. We don't know how they work.

Prompt engineering at this point is basically only experiment, error, repeat.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: