manton
manton

Dan Murrell blogs about how he uses ChatGPT to add features to his iOS app:

Why should I spend time writing all the code when I could instead describe what I want, and evaluate and adjust minor bits instead?

This is going to become more and more common. Importantly, you can’t blindly follow AI coding advice. A human still needs to be in charge who understands the code.

|
Embed
Progress spinner
numericcitizen
numericcitizen

@manton how many dev will accept the code without fully understanding what is going on? Is asking ChatGPT to explain enough?

|
Embed
Progress spinner
manton
manton

@numericcitizen I'm sure some devs will run with code they don't understand, but I wouldn't. Gotta understand at least the basics to be comfortable with shipping it. Or test really, really well.

|
Embed
Progress spinner
7robots
7robots

@manton FWIW - I've found it most useful to create a first approximation of a specific function I might need. I'll then go in and modify or adapt the AI-generated code to my purpose. I tend to keep the unit at the "block of code" level, rather than an entire program. This approach has been pretty good at helping me avoid blind adoption. I've also found it moderately useful at finding bugs or logic errors.

|
Embed
Progress spinner
In reply to
KimberlyHirsh
KimberlyHirsh

@manton This is true with audio transcription, too.

|
Embed
Progress spinner
manton
manton

@7robots This is mostly my approach too. Helping implement fairly narrow functions. That it works at all on more broad problems is fascinating to me, though.

|
Embed
Progress spinner
manton
manton

@KimberlyHirsh Agreed. You really need a human to review and edit. Transcription is so good now, but there’s always that last 1% that might be off.

|
Embed
Progress spinner