“Prompt engineering” is such a bizarre line of work. You’re trying to convince a machine trained on a huge pile of (hopefully) human-generated text to produce some useful output by guessing what sequence of human-like words you must put in to make it likely that the model will produce coherent, human-like output that is good enough to pass downstream.
You really have no idea how your prompt caused the model to produce its output (yes, you understand its process, but not the actual factors that contribute to its decisions). If the output happens to be good, you still have no idea how far you can push your input before the model returns bad output.
Prompt engineers talk to the model like a human, because that’s the only mental model they have for predicting how it will respond to their inputs. It’s a very poor metaphor for programming, but there is nothing better to reach for.
I’ve heard that prompt engineering is a way to talk to a computer like a human, and not through an API. But the reason APIs exist is because programming problems require a degree of PRECISENESS that is not available through human language. Programming languages that try to emulate human language (e.g. COBOL, AppleScript) quickly run into situations where human language is too imprecise or verbose to express the complexity and precision needed for actual programming; their grammars develop weird and stilted corners to accommodate this un-human degree of precision.
But we’re going to Prompt Engineer our way into the future, huh? Yeah, sure.
@drahardja The ‘guardrail’ preamble prompts crack me up. Nothing like pure English rules definitions in an unspecified format and zero binary compatibility (whatever the equivalent is)
@bgannin The more I look at actual prompts being input into LLMs on which literal billions of dollars of investments (and millions of dollars of daily running costs) rely on, the more incredulous I become.
@drahardja It was circa 40 years ago that I, as a kid, attended my first computetr conference. I still remember a booth that was selling a solution where you could write your needs in "natural" language, making a programmer unnecessary. The dream lives on, as do the credulous people paying for "solutions".
@drahardja No. Because LLMs aren't yet "technologies", so much as phenomenon. They are things we've learned to make that we can coax interesting responses from - and sometimes they can even be useful, though frustrating and require a lot of coaxing to get the responses we're hoping for. They're things we can experiment with and play with, but they aren't tools yet - because we don't have simple straight forward ways to get them to do what we want or understand what it is that we want.. yet.
@drahardja semi serious question, can the engineering profession get these folks to either back off or to hold their work to engineering standards (lol)? It seems about on the level of horoscope engineering.
@drahardja this seems odd to me. Like I don’t really use these tools nor do I necessarily think “prompt engineering” will be around forever, but there are lots of things in life that are more about whispering to semi-wild systems that can be useful. I don’t think that having total control or even understanding is as important as usefulness in many context. It does introduce considerable systemic risk, I think, but I think there’s a common conflation when people talk about AI.
@drahardja the very existence of this need is a huge clue to the genuine maturity of the technology. Hint: it is not a genuine intelligence, it’s a pile of language statistics that has to be coerced into sounding intelligent and useful by extremely motivated humans pulling the puppet strings.
@drahardja Super strange to think that before prompts the equivalent was knowing how to do an advanced search with the right words to find the answer to a problem.
Add comment