4 ms·

Am I the only one appalled by the audacity to call the writing of vague instructions for an AI "engineering"? However many levels of indirection there might be

Am I the only one appalled by the audacity to call the writing of vague instructions for an AI "engineering"? However many levels of indirection there might be at play.

jojobas 3y ago
If social engineering counts, this should too.
- bertylicious 3y ago
  Fair point. Maybe the term engineering has a tradition of being used in contexts that are not associated with hard science. Another example: the term "software engineering" was invented to point out the lack of scientific foundation for programming practices and to motivate the community to fix that. Or maybe it's just me associating engineering with hard science.
smokel 3y ago
This made me look into the etymology of "engineer", and I was slightly surprised that it is all about weapons, and not so much bridges.
muzani 3y ago
I see two definitions of engineering: 1. Building engines from scratch, i.e. something that converts a thing from X to Y consistently. 2. Knowing enough about an engine to get it to do what you want it to do, as well as maintaining it. Prompt engineering initially started from the latter, back before ChatGPT and stuff which you'd just give instructions to. Engineers have to be familiar enough to know how it works, what it's strong at, what it isn't, things like margin of failure, all that stuff. A lot of it is just prompt alchemy, I guess, but the lower part of this article has some juicy reverse engineering.
intelVISA 3y ago
As opposed to writing vague instructions for a compiler?
- bertylicious 3y ago
  How do you mean? Do you mean source code? What's vague about that?
  - pama 3y ago
    Except when writing in assembly, the compiler can make unintuitive decisions for how to transform our source code into something that runs on a particular hardware and as modern hardware complexity increases it becomes unlikely that a human can guess the exact code that will run. LLM is a higher level of abstraction than most existing programming languages, but the analogy of the GP is meaningful to me.
    - kangalioo 3y ago
      Good point. When I'm going back and forth writing code and checking its assembly, I also reason about the compiler almost like an intelligent being: "oh, it doesn't see this condition, so if I help it a little bit, then... Ah now it unrolled, and inlined, but accidentally trashed instruction cache, how will I convince it not to..."

Am I the only one appalled by the audacity to call the writing of vague instructions for an AI "engineering"? However many levels of indirection there might be at play.

jojobas 3y ago

If social engineering counts, this should too.

bertylicious 3y ago
Fair point. Maybe the term engineering has a tradition of being used in contexts that are not associated with hard science. Another example: the term "software engineering" was invented to point out the lack of scientific foundation for programming practices and to motivate the community to fix that. Or maybe it's just me associating engineering with hard science.

smokel 3y ago

This made me look into the etymology of "engineer", and I was slightly surprised that it is all about weapons, and not so much bridges.

muzani 3y ago

I see two definitions of engineering: 1. Building engines from scratch, i.e. something that converts a thing from X to Y consistently. 2. Knowing enough about an engine to get it to do what you want it to do, as well as maintaining it. Prompt engineering initially started from the latter, back before ChatGPT and stuff which you'd just give instructions to. Engineers have to be familiar enough to know how it works, what it's strong at, what it isn't, things like margin of failure, all that stuff. A lot of it is just prompt alchemy, I guess, but the lower part of this article has some juicy reverse engineering.

intelVISA 3y ago

As opposed to writing vague instructions for a compiler?

bertylicious 3y ago
How do you mean? Do you mean source code? What's vague about that?
- pama 3y ago
  Except when writing in assembly, the compiler can make unintuitive decisions for how to transform our source code into something that runs on a particular hardware and as modern hardware complexity increases it becomes unlikely that a human can guess the exact code that will run. LLM is a higher level of abstraction than most existing programming languages, but the analogy of the GP is meaningful to me.
  - kangalioo 3y ago
    Good point. When I'm going back and forth writing code and checking its assembly, I also reason about the compiler almost like an intelligent being: "oh, it doesn't see this condition, so if I help it a little bit, then... Ah now it unrolled, and inlined, but accidentally trashed instruction cache, how will I convince it not to..."