Hacker News Reader

Comment by A4ET8a8uTh0 | Hacker News Reader

I am not sure that can be done that way. There are issues that are culture or specific business need related. Bad requirement is the hard part it seems. I think we all have been in situations, where 'common sense' is not common ( and does not translate to the same thing to various layers for people who pass it; and that does not even account for translation ).

This is usually how we end up with weird situations like currency coded as text so it can't be sorted by amount by end user and so on.

I agree that specs make or break the project, but at what point is it ok to assume "x should do y".

View on HN

Yeah, on top of that add wading through legacy code, figuring out why wtf, this is crap, just to figure out most times it is actually crap, but then at least equal amount of times that there was a very good reason to do it that way, either technical or by business logic... cannot imagine how any AI system could have success there without getting towards the true AI singularity.

This kind of code that I'd believe makes up 99% of all code out there that is worked on, is not available to the public for training. What to do about that? Rewrite all?

I think this may be the case, the AI will figure out what the legacy code does and then completely re-writes it using whatever modern standards and ability to host it all in the cloud.

The stochasticness probably gets you into trouble here. If you get different results each time you try to "figure out and rewrite" the legacy program... you're gonna need to be able to right some extremely thorough, perfect test cases. And how much of that are you willing to put to chance?

If it's not trained on the legacy code, I have my doubts it will be able to re-write it well enough to replace a human baby sitting it, particularly when we're talking about large projects. As impressive as ChatGPT is at spitting out code examples, what gives you confidence that and LLM could rewrite a million lines of legacy code it has never seen before? That seems to be well beyond the ability of any AI that's been developed so far.

Reply on news.ycombinator.com