So like when it's as good as it'll ever get (I think it'll hit a wall maybe not even that far into the future)
It'll get you 90% of the way there and then need a myriad of fairly complicated issues solved by a human, and that human programmer will have to be pretty exceptional so that they can read badly organized not very readable code and do it in an amount of time that makes sense for this procedure to even be a more efficient way of doin things
2 Name: Apned2024-07-10 00:08
I'm a Denpa
3 Name: Anonymous2024-07-10 04:17
every LLM use case for programming that actually works more than 10% of the time boils down to stuff we could do without LLMs for less than 1% of the cost. But now companies are giving Microsoft a penny every time some guy says "generate the setter for this variable" as if every IDE ever couldn't do that for free (and more consistently correctly).
Everytime I hear of coworker bragging about how they used copilot to generate a skeleton or rename variables, I realize that LLMs are not making good programmers better. They're making people who have no business being anywhere near programming believe they're good enough to be allowed near code, and good programmers are going to be inundandated with the tasks of reviewing and fixing broken, AI-generated code with useless AI-generated commits messages, where the original author has no idea what's going on.
It'll get you 90% of the way there and then need a myriad of fairly complicated issues solved by a human, and that human programmer will have to be pretty exceptional so that they can read badly organized not very readable code and do it in an amount of time that makes sense for this procedure to even be a more efficient way of doin things