Daniel JalkutWe're at a turning point, where everybody 20 years old or so, and older, will have a fundamental advantage against every younger programmer when it comes to "coding with AI". Younger programmers will have no natural incentives to learn non-vibe coding techniques. Things are gonna get weird.
Parallels to the past: C Programmers who benefited from knowing Assembly. Java programmers who benefited from knowing manual memory management. UIKit programmers who benefited from knowing AppKit. SwiftUI programmers who ... well, they kind of got a raw deal.
Mike Sax@danielpunkass Will it matter? I have no idea how to make a punch card, and it hasn’t limited my work.
Thomas Sturm@mikesax @danielpunkass It would matter, if every OS would still be using punch cards under the hood.
Young developers that now create some app with an LLM will be out at sea as soon as a deeper integration with the OS is necessary. Also, they will have learned none of the underlying skills that would allow them to build a career in this field.
What are they going to do in the future - forever just ask an LLM to please do the coding for them?
Matthew Haughey@danielpunkass I feel like people 15 years younger than me were all pushed into pair programming with another person that might do half or more of the work, maybe vibe coding is just pair programming but with a robot?
Scott Wright 🏳️🌈🏳️⚧️@mathowie
The problem, right now, is thinking that the AI half of the pair is the wise, experienced one. It's fast, but often chooses a bad path quickly.
Even if both of the pair were inexperienced, they talked through the problem and often got to an efficient solution. If you assume the AI is all knowing (not just fast) and don’t inspect the methods, you’ll accept bad code.
@danielpunkass
The problem, right now, is thinking that the AI half of the pair is the wise, experienced one. It's fast, but often chooses a bad path quickly.
Even if both of the pair were inexperienced, they talked through the problem and often got to an efficient solution. If you assume the AI is all knowing (not just fast) and don’t inspect the methods, you’ll accept bad code.
@danielpunkass
draNgNon@danielpunkass I don't quite think that's the same, becuase the models used will be biased towards older stuff they are trained on, and thus necessarily lagging in providing the most recent innovations.
@draNgNon maybe, but the models are trained more and more to do their own research. They search the web way faster than I can, and bring me up-to-date advice.
Matt Massicotte@danielpunkass yes and I do not see how this phenomenon would be limited to programming
Dave Reed@danielpunkass As someone who teaches computer science, I worry that AI is preventing stuents from learning the core concepts that are necessary to understand, even when using AI later in their educational and work careers. Some students see this and are avoiding AI for assignments (but still using it for reviewing for exams, etc.) while others are fine with AI doing their homework that is meant to develop their own skills.