> Very hard for humans to keep up with LLM-generated code
Not at all. It's hard to keep up with lines of code per day, i.e. quantity.
It's not difficult for humans to keep up with the quality, and produce far superior quality code by writing it themselves or instructing an LLM what exactly to implement, down to each function's API surface (if typing is the bottleneck).
A lot of you people are forgetting that there's an engineering aspect to programming, and LLMs are really really really terrible at it out of the box.
Let's say you can let AI design and 3D print 1 large real-life bridge per week. Would you say that humans can't keep up? Or would you point out that there's a difference between what civil engineers do and what an LLM with a 3D printer does?
Not at all. It's hard to keep up with lines of code per day, i.e. quantity.
It's not difficult for humans to keep up with the quality, and produce far superior quality code by writing it themselves or instructing an LLM what exactly to implement, down to each function's API surface (if typing is the bottleneck).
A lot of you people are forgetting that there's an engineering aspect to programming, and LLMs are really really really terrible at it out of the box.
Let's say you can let AI design and 3D print 1 large real-life bridge per week. Would you say that humans can't keep up? Or would you point out that there's a difference between what civil engineers do and what an LLM with a 3D printer does?