Well, nobody cares about you're views about stickers and tattoos, but you still commented. Compared to you though, nobody here called you a retard; maybe that says something about how people are here or how you tend to use inflammatory language when it's not needed.
I care! I dunno if I agree with him about all tattoos (surely some are tasteful even if it's a minority), but stickers are definitely all tasteless and kitschy.
It can benefit from more complex algorithms, but I would stay away from "AI" as much as possible unless there is indeed need of it.
You can analyse your data and make some dynamic thresholds, you can make some small ML models, even some tiny DL models, and I would try the options in this order.
Some cases do need more complex techniques, but more often than not, you can solve most of your problems by preprocessing your data.
I've seen too many solutions where a tiny algorithm could do exactly what a junior implemented using a giant model that takes forever to run.
Would that help that much nowadays when peer-reviewing is just 2 students in a trench coat?
In my opinion, especially on more novel works like this, the discussions between people in the field are way more useful than a stamp of approval from whatever journal.
It still is super relevant. Most computer vision done outside academia is still based on older stuff, or classical computer vision algorithms. You don't really get so many chances to use the latest models and techniques, as most often than not, they are not that relevant, or are only for extremely specific cases, or you just don't need something that complex.
I want to delve into some more technical interpretations in the future about monosemanticity, the curse of dimensionality, and so on. Although I worried that some parts might be too abstract to understand easily for non-technical people, so I wrote a quick intro to ML and encoders as a stepping stone to those topics.
Its purpose is not necessarily to give you a full technical explanation but more of an intuition about how they work and what they do.
Thought it might be helpful to some people here as well who are just getting into ML; hope it helps!
I've only quickly skimmed the course on fast.ai and it seems to me that it has a more practical approach. It also says that math is not really important and high school math should do, which i do not agree with unless you only care about how to implement different techniques and hope for the best.
To understand the subject properly, I would say that you actually need a bit of a mathematical background to be able to understand the whys and hows of ML and DL.
For this, I would start with Elements of Statistical Learning [1] to get all the background.
After that, probably Andew NG's Coursera Course gives you the next steps and also how to implement all those methods. Maybe in the meantime there are better courses, but I'm not sure, I haven't looked into it in a long time.
The Deep Learning book [2] is good to get you started on all the classic DL methods, but won't really cover anything new (like vene transformers which by now are not even new anymore).
After that, you can probably start looking into trying out projects by yourself, even going on arXiv to check out newer research and familiarize yourself with the academical lingo and way to describe the maths and practicalities (imo could be way better, but academics have a big ego they cannot leave to deflate).
Also, might be important to mention that this whole thing set up like this, and made properly would take 1+ years with bachelor's knowledge already, so if you just want some basic knowledge, you could skim through some of those, but this would be the path I would recommend to get proper knowledge of all the bases, and fully understand what's happening in the field.
I always liked the idea of taking notes on paper, but honestly, after uni, I kind of stopped taking them in that format for personal notes as I felt like I never found whatever I needed when I needed it.
In the end, a mix of digitally hadwritten notes in Nebo, and logseq seemed to work best for me.
My mom used to tell me when I was young to never run after money and choose a career in which I could do something that made me happy. As working on something you enjoy means you'll do it with passion, that will be seen by others in the quality of your output, so money (or fame, or whatever you're after) will come naturally.
I was around 5, and video games was what made me happy. XD
As I grew up and learned more about the field, I figured it's not necessarily video games that I want to make, but definitely software science.
I ended up in electronic engineering because I was curious about how computers work at a low level.
After that I ended up in ML/DL/data science because it looked cool and wasn't that hard to understand for me as I had all the math prerequisites from the EE degree.
In conclusion, I would say that, for me at least, the path came quite naturally, and every step I took in my career made sense at the time. I never tried to really force a long-term plan on my career, so whenever a move or a change felt natural for what I wanted, I made it. I would say I'm pretty happy with what I'm doing now, so at least it worked for me.
There are two schools of thought. 1. Follow the money. 2. Follow your passion.
It's hard to know which is right (and I suspect both are), because my passion and the money are in the same place (tech).
Which means folk in tech (myself included) can successfully anecdote either branch.
Personally I probably land on the passion side, but I you're passionate about archaeology, well, good luck to you.
To the original question I say - be curious about everything. Try and learn from those around you. Try and add value outside of just support. Expand your support skills outside your specific job. Learn about the OS, or SQL, or coding.
You don't need to map out your niche just yet. Take some time to be a generalist with a wide experience. From there you'll likely find yourself gravitating to a field.
I've been taking a look recently at Centsible and planning to try it out.
It's similar to the YNAB methodology, only on android or ios for now, but honestly that's what i've been looking for. It doesn't have all the fancy features that other apps have, but probably that's why it is much cheaper as well.
Copilot looks pretty good, but it's ios only so I wasn't able to try it out.
I believe the math background is quite important in ML, and having a good understanding of it from the beginning helps tremendously.
You can check my answer from before here:
https://news.ycombinator.com/item?id=36477464
