[esperent]

> Why can't I rerun the prompt that resulted in your work next year, when models have gotten better?

Because you could also just point the better model at the generated code and tell it to improve it, so why save the prompt too?

[Zambyte]

Because improving a high schoolers project is probably not as good of an idea as giving the original task to a senior engineer.

[bluGill]

If the high school project has been improved by many engineers over a few years it likely is complex enough that a senior engineer cannot rewrite it for a reasonable cost. It isn't clear if next years models will be enough better that they can rewrite it for a reasonable cost. If they are they can probably extract the requirements and special cases from the code.

[Zambyte]

Following my analogy, the high school project would have been continuously extended by high schoolers passing the project on from class to class each year, rather than "improved by many engineers". The "engineers" (future models) don't get the project until currently models have had their way for a while. I think that makes the "rewrite from scratch" plan a whole lot more compelling.

[esperent]

It's a bad analogy. Say what you will about the strengths and weaknesses of current LLMs but there's no "high-schoolers" (outside of rare prodigies) who can write code at the level of a frontier LLM.

[Zambyte]

High schoolers are to senior software engineers as current models are to future models. Is that really so hard to follow?