Hacker News Reader

Comment by Towaway69 | Hacker News Reader

Oracle Oriented Programming, ask the oracle and it shall answer.

The whole AI hype reminds me of the ancient Greeks going to the Oracle at Delphi to obtain answers.

From the perspective of an outerspace alien, what is the difference between asking a god like figure for answers or an AI is some data center?

View on HN

I'm kinda the Oracle at my workplace. People come to my desk to obtain answers and I google for a couple minutes and deliver them.

I thought maybe ChatGPT would replace me, but so far the same people come so I can prompt it for them.

If your relationship to the subject you're asking about is equivalent to an oracle asking an inscrutable deity, then I don't know that you're going to get much value out of it. OTOH, if you know enough about the subject where the response is something you can evaluate for whether it actually makes sense, and can take and edit or integrate in a useful matter, you'll likely find that the LLM's provide a very helpful speedup that boosts productivity. At least that's been my experience.

Stackoverflow is the same: use the code if you understand it. But with AI I have the fear people won't take the time to understand the code - copy, paste and it's done, after all it comes from AI!

How maintainable is software that was not fully understood when it was built? When it breaks in 6 months because some api changed. I guess the solution will be to get AI to fix it ... Sigh.

Industrial-scale copy-pasta is definitely a concern I have around LLMs, though in practice I'm not sure it'll actually be _so_ much worse than it is now. Consider, for instance, one team that aims to have good test coverage and productive code reviews, and one that doesn't, and assume that on both teams there are some developers who carelessly copy-paste whatever the LLM spits out. I suspect that the first team will still end up doing pretty well, whereas the latter will get more terrible. So one outcome might be a greater delta between teams these two types of teams, with the takeaway being that in the new AI world, practices around code testing and reviewing are more important than ever.

Reply on news.ycombinator.com