Hacker News new | ask | show | jobs
by smoyer 145 days ago
When I was in high school, we were not allowed to use calculator for most science classes ... And certainly not for math class. I'm ten years, will you want to hire a student who is coming out of college without considerable experience and practice with AI?
6 comments

LLMs work best when the user has considerable domain knowledge of their own that they can use to guide the LLM. I don't think it's impossible to develop that experience if you've only used LLMs, but it requires a very unusual level of personal discipline. I wouldn't bet on a random new grad having that. Whereas it's pretty easy to teach people to use LLMs.
If all they know is AI, and they supplanted all their learning with AI, why even hire them? Just use the AI.
Kids need to learn the fundamentals first and best. They can learn the tools near the end of school or even on the job.

I loved computer art and did as many technical art classes at university as I could. At the beginning of the program I was the fastest in the class, because we were given reference art to work from to learn the tools. By the end of the class I couldn't finish assignments because I wasn't creative enough to work from scratch. Ultimately I realized art wasn't my calling, despite some initial success.

Other kids blew me away with the speed of their creations. And how they could detach emotionally from any one piece, to move on to the next.

Why would I want to hire such a student? What makes him better the better pick than all the other students using AI or all the other non-students using AI?
Yes, it is much easier to train someone to use AI than to train them to have sufficiently baked-in math and language skills to be able to leverage the AI.
Should I, by some miracle, be hiring, I'd be hiring those who come out of college with a solid education. As many have pointed out, AI is not immune to the "garbage in, garbage out" principle and it's education that enables the user to ask informed and precisely worded questions to the AI to get usable output instead of slop.