[GianFabien]

Skate to where the puck will be (Wayne Gretzky). AI is backward looking (where the puck was) technology. It is trained on what is already widely known. In experiments it handles bar exam type questions well. But not systems engineering questions that require understanding the problem domain.

Of course, you could develop foundational IT knowledge: at least one programming language, an OS, a framework or two. But the key to long term success is becoming wickedly knowledgeable about some problem domain, e.g. biotech, some niche in finance, supply chain, medical analysis automation, etc.

Once you establish core competence in the domain of your choice, your future IT learning will be directed by the needs of the problems you are solving.

[subtra3t]

> Of course, you could develop foundational IT knowledge: at least one programming language, an OS, a framework or two. But the key to long term success is becoming wickedly knowledgeable about some problem domain, e.g. biotech, some niche in finance, supply chain, medical analysis automation, etc.

What do you think is the best way to do this? Or do you think that the best most "efficient" way differs for each domain?

[GianFabien]

Perhaps you are too focused on the ideal of "efficiency". Seems like another form of procrastination.

Just get started. The Richard Feynman technique is very productive. Ask a question, and then research until you answer it. Along the way you will ask many other questions. Once you answer the first question, pick the most interesting new question and repeat. Before you know it, you will be more knowledgeable in that area than the average.

AI can be a great assistant. It will do the legwork for you. But you need to verify to ensure that you are not fed hallucinations or erroneous information. With increasing knowledge and critical thinking you will get better at panning for knowledge gold.

[subtra3t]

Thanks!

[muzani]

Straight example: I use Android with Kotlin and Jetpack Compose. Both are amazing tech, cuts down LOC by a third, reduces complexity for large codebases massively.

ChatGPT is familiar with all of these techs. It has read the fucking manuals and oh god most people don't read them because they're not written for humans. This is the perfect environment for AI to thrive.

But it'll write code based on similar code around 2019 or so. A lot of code out there is terrible. Much of it was written by contractors who don't want to get the work done. These guys are happy to see complexity go up exponentially. Many people work as contractors and in agencies, quit, do their own thing, adopt shitty architectures that were worse than no architecture.

So we see applicants submitting this crappy code that they do understand, but we ask them why they picked this over the fancy new stuff. Most of the time they didn't pick anything. They told AI to do it, and AI went to where the puck was.

I mean I train AI to write the fancy new code. But you have to know what it should look like.