I'm currently in first year, but I'm looking forward to specializing in AI or formal methods. At the moment, statistical learning and mathematical logic are, naturally my main subjects of extracurricular study.
While you can look at my projects to see what I'm currently experimenting with (currently mostly whimsical thoughts I have), currently most of my more ambitious ideas have been left on the backburne, among which are a cluster made from 80s 8-bit microprocessors, a classical AI spell-checked and grammar analyzer, and a classical AI system for document segmentation and markdown extraction from pdf.
Especially since taking AP courses, there has been a lot of focus from my peers on optimizing studying and grinding out as much work as possible, whether it be personal projects, extracurriculars, or academics.
I'd certainly be lying if I said that I don't fall into this trap myself. When you sit down and get into a problem set or a cool programming project or learn about some new idea, it's easy to get so caught up in what you're doing that you forget to take care of yourself. And when it's not, it's easy to blame yourself and to make plans to maximize productivity for the remainder of the day.
While I think that certainly an emphasis on efficiency is good and it drives you to produce
better work and to hone your skills, it can also be quite toxic when you inevitably at some
point fail to meet your expectations and beat yourself up when reasonably, you can't
While it might sound whimsical and sweet and completely ignorable, taking care of yourself
is probably one of your top priorities. If you need a reason, it's because you won't be
able to cherish the fruits of your success and all if you're dead at age 19 of stress and
cardiovascular disease. If that doesn't convince you, taking a break and letting your
brain subconsciously toss around the concepts you've been grinding at can be just what you
need. Case in point: Albert Einstein conceived some of his most revolutionary ideas being
bored as a clerk, not grinding away at his studies. So overall, I think that
Lately, most people around me have been using LLMs. Where I study, it's almost the norm
at this point—if you can't solve a problem, you ask AI. If the AI can't solve it,
you ask another AI. For perhaps less educational purposes, it seems that a lot of people
also use it to hack together personal projects and to get away with doing less in group
projects. To be fair, I think this totally works for most of my peers since
(most of them) still do emphasize
I've made the conscious choice to avoid AI, at least for now. I recognize the potential it already has for streamlining workflows and pruning out repetitive, well-defined tasks, but I want to avoid becoming increasingly dependent anytime I don't want to do something properly. I feel that especially at my stage of learning, going through and solving problems, no matter how repetitive, manually helps build valuable intuition. Thus, until circumstances change, I have decided to stick to reading documentation and building basic skills.
I think that Neural Networks are an impressive technology, and it's hard to cast doubt on the fact that they've achieved some very impressive results already with LLMs, image generation, and a number of other fields where the data is too unstructured or ambiguous for traditional algorithms to analyze. I think that the technology right now could be overhyped, but it doesn't take away from the sheer brilliance of the technology.
Nevertheless, I prefer not to trust AI. Although LLMs have gotten shockingly accurate in the last year or so (the free GPT offering has been outdoing most of my class in Calculus), they're still known to fail, especially when doing long, drawn-out numerical computations. This unfortunate tendency, combined with the fact that it's difficult to verify the logical and factual "correctness" of an answer (the chain of thought reasoning helps here but is imperfect), make using AI a slight gamble.
This limitation of current LLMs is part of the reason why I'd like to continue studying formal logic! While classical AI is certainly not nearly as capable as a GPT in many tasks, it also provides strictly logical and traceable reasoning for its conclusions. Besides, I think it would be really cool to have a mostly complete theoretical understanding of the mathematical properties of an AI algorithm.