3 ms·

English, and other languages, are vague and imprecise. I've never understood why folks think they can write code "more efficiently" with a prompt rather than co

by danjl 2y ago

English, and other languages, are vague and imprecise. I've never understood why folks think they can write code "more efficiently" with a prompt rather than code? Are people willing to give up control? Let the LLM decide what is best? The same is true for generative art -- you get something, but you only have marginal control over what. I think this will always be something that is useful for the simplest things, simplest apps, simplest art, etc.. A race to the bottom for the bottom of the complexity stack. As problems become more complicated, it would take a great deal more prompt language to specify the behavior than code.

emseetech 2y ago
I wonder about this as well. Go has 26 keywords. English as 170,000 active words. I don't want to code in English.
- BeetleB 2y ago
  Binary has only two bits. I really don't want to code in two bits.
  - emseetech 2y ago
    There's a sweet spot for identifiers when writing software and it's somewhere between 2 and 170,000 but likely closer to 2.
- Cordiali 2y ago
  Most of those words are engPM packages. The trick is to not include the `architecture` package, if you're just making toast.
throwup238 2y ago
For most things I approach it from both sides: I give an English prompt but also provide some interface, function signatures, or even just a return type that constrains the LLMs output. For example, I frequently have to implement Serde's traits for some serialization format or to marshal types, most recently to translate types from Rust to Qt QML's Javascript. By giving it some context (Serde traits and QT docs) I managed to do it with Claude in about an hour, which is roughly how long it would have taken me just to get up to date with the documentation if I had tried it myself.
kmoser 2y ago
Nobody in their right mind gives up control or lets the LLM decide what's best. They take the output of the LLM as a starting point, changing it where necessary to accomplish their goals. This is no different than starting with boilerplate code, e.g. from a tutorial or manual or other project (or, heaven forbid, an Internet search), and changing it for your specific needs. As for imprecision, a developer is perfectly capable of writing imprecise, ambiguous, or just plain buggy code by themselves. If you think your code is better than that produced by an LLM (and I'm not saying it isn't), then by all means use it. But the fact is that an LLM, in some cases, can produce code better and/or faster than it would take some people to write themselves who don't want to spend several minutes or hours stumbling through finding the exact syntax or algorithm, let alone Googling for a good starting point. As for the non-determinism that many people are decrying, this is no different from Googling and finding several examples, each of which is written for a particular use case, none of which ever seem to match yours exactly. Caveat emptor if you simply cut and paste without reading, just as with code produced by an LLM.
- danjl 2y ago
  As I said, the LLM will be a good replacement for the simplest cases. The behaviors you describe are far more common among junior programmers working on simple apps.
  - BeetleB 2y ago
    > The behaviors you describe are far more common among junior programmers working on simple apps. Or among expert programmers who need to code in something out of their domain for a small portion of their job. Suppose you suddenly are required to write a VBA macro for Excel for your job. It's a one off task - not something you'll do repeatedly. Do you prefer learning VBA for Excel and crafting a solution or asking the LLM and verifying it's solution by looking at the docs? Hint: If you use the macro recorder in Excel and inspect the code you are closer to the LLM end of the spectrum.
- fuzztester 2y ago
  this is one of the most perspicacious comments I have ever read about llms, with the disclaimer that i have not read a large number of comments on the subject. I have read some however. however, referring to the first paragraph of the comment above, the problem is that many people are not in their right minds :)

by danjl 2y ago

emseetech 2y ago

I wonder about this as well. Go has 26 keywords. English as 170,000 active words. I don't want to code in English.

BeetleB 2y ago
Binary has only two bits. I really don't want to code in two bits.
- emseetech 2y ago
  There's a sweet spot for identifiers when writing software and it's somewhere between 2 and 170,000 but likely closer to 2.
Cordiali 2y ago
Most of those words are engPM packages. The trick is to not include the `architecture` package, if you're just making toast.

throwup238 2y ago

For most things I approach it from both sides: I give an English prompt but also provide some interface, function signatures, or even just a return type that constrains the LLMs output. For example, I frequently have to implement Serde's traits for some serialization format or to marshal types, most recently to translate types from Rust to Qt QML's Javascript. By giving it some context (Serde traits and QT docs) I managed to do it with Claude in about an hour, which is roughly how long it would have taken me just to get up to date with the documentation if I had tried it myself.

kmoser 2y ago

Nobody in their right mind gives up control or lets the LLM decide what's best. They take the output of the LLM as a starting point, changing it where necessary to accomplish their goals. This is no different than starting with boilerplate code, e.g. from a tutorial or manual or other project (or, heaven forbid, an Internet search), and changing it for your specific needs. As for imprecision, a developer is perfectly capable of writing imprecise, ambiguous, or just plain buggy code by themselves. If you think your code is better than that produced by an LLM (and I'm not saying it isn't), then by all means use it. But the fact is that an LLM, in some cases, can produce code better and/or faster than it would take some people to write themselves who don't want to spend several minutes or hours stumbling through finding the exact syntax or algorithm, let alone Googling for a good starting point. As for the non-determinism that many people are decrying, this is no different from Googling and finding several examples, each of which is written for a particular use case, none of which ever seem to match yours exactly. Caveat emptor if you simply cut and paste without reading, just as with code produced by an LLM.

danjl 2y ago
As I said, the LLM will be a good replacement for the simplest cases. The behaviors you describe are far more common among junior programmers working on simple apps.
- BeetleB 2y ago
  > The behaviors you describe are far more common among junior programmers working on simple apps. Or among expert programmers who need to code in something out of their domain for a small portion of their job. Suppose you suddenly are required to write a VBA macro for Excel for your job. It's a one off task - not something you'll do repeatedly. Do you prefer learning VBA for Excel and crafting a solution or asking the LLM and verifying it's solution by looking at the docs? Hint: If you use the macro recorder in Excel and inspect the code you are closer to the LLM end of the spectrum.
fuzztester 2y ago
this is one of the most perspicacious comments I have ever read about llms, with the disclaimer that i have not read a large number of comments on the subject. I have read some however. however, referring to the first paragraph of the comment above, the problem is that many people are not in their right minds :)