Hacker News new | ask | show | jobs
by BadHumans 828 days ago
Are you actually learning the concepts though? I'm not trying to be inflammatory but a lot of people I see using LLM use them as a replacement for StackOverFlow. They aren't learning anything they just need it to spit out code that works.
3 comments
drusepth 828 days ago
Yes, absolutely. AI has been an amazing learning tool for me as I transitioned into gamedev working at a Unity studio. What used to take upwards of hours to research or debug can now be solved in minutes by asking AI about solutions, best practices, and feedback on code.

I am constantly using AI to learn more every day -- and learning at a faster rate than I was just pouring through the internet for those same answers.

link
bongodongobob 828 days ago
Absolutely, and I get the sentiment.

I'm not just using it to generate code, but asking specific questions. Even really basic stuff like, write me a Fibonacci number generator in X language. Helps me understand the syntax immediately. I can start from a place of familiarity. A lot of times when picking up a new language, the hello world stuff is kind of drawn out. Give me a program in Rust that generates 50 random numbers and sort it from smallest to biggest. Give me the basic framework for a Flask app, etc etc.

Or, I have an app that scrapes a website. What is the best architecture to put this into a database to reduce scraping the same thing multiple times?

I use it like a project manager/software architect/tutor in that order I think.

It massively reduces the friction to just getting things going.

link
avery17 828 days ago
Do you need to really learn if it works and ships?
link
happytoexplain 828 days ago
This isn't a commentary on AI, but specifically the idea that all you need to achieve is "it works". That's based on the fallacious premise that a human can reasonably bucket a program (especially one as complex as an end-user application) into "working" and "not working". You don't know it's "not working" until you know, which may in some cases be never.

As for AI specifically, robust testing is already difficult with human-curated logic. It's potentially more likely to miss failing cases, or non-binaries (e.g. performance, security), when testing genAI. More subjectively, your conjecture represents a mindset that is demotivating to people with certain flavors of creativity/passion. Also, overreliance on AI may blunt skills that may remain useful, and even important - i.e. it may exacerbate an expertise scarcity.

(this may make me sound more negative on AI that I am - remember that I'm commenting in the context of the post you're replying to)

link