Has anyone moved into A.I. programming from a decently long web dev career. Or maybe you work in AI and you have 2c.

Worried about what comes next. So, I've decided to pivot hard into core math and comp sci like linear, calc, probability, dsa.

Worth it? Naive? I was a calculus tutor way back.

I’m working on doing the same. I think it will work out, but actual ML work usually requires you to do research, and companies often want someone with a masters or PHD for that stuff.

Worst case scenario, the math and comp sci skills are always useful.

I think I will try and learn enough pattern matching to support our robots orienting themselves to our target welds. Or at least understand what other companies have done.

I mean, I have never really moved into nor out of AI.

AI, when not on the bleeding edge, is very basic math. (Sure, this might come from me being a physicist, but still, it's mostly basic statistics).

You don't need to understand how bleeding edge models work, but having some high level notions for the concepts will go a long way because whenever you have to dig deep you'll find it's not so alien.

What do I mean by this?

Know what the problems are.

Classification? Know what a SVM is, know what a multiclass is, know what KNN is. Not *how* they do it, *what* they do.

Recommendation? Know what embeddings are, what vectorial databases are, etc.

LLM? They are basically glorified recommendators.

Knowing what they do (and what they *don't*) means you now know what solution fits your actual problem, and can get many off the shelf solutions (just like in web dev now).

@CoreFusionX I checked out a AI 3D model generator website. I old it I wanted a female model. It kept generating Lizzo. So I know which AI services I don't want.

@Demolishun

Hahaha, the problem nowadays, like in every "gold rush", is all the tech vampires that smell the blood and want to get rich by just selling a wrapper for existing tech.

Which, mind you, is something all agencies do, and not inherently wrong, but on a gold rush market, shit offerings thrive.

They'll die out on their own.

How? By doing what I said.

If you can be the guy in your company that knows you just have to fire up opencv and write 100 lines of python, or fire up TF and write 100 lines of python (why do they all love python?), your company will see they don't need to pay a crapload of a subscription for a shitty service (and hopefully give you a well deserved raise based on the money you saved them).

@CoreFusionX thanks Core for some insight. In addition to the maths (for me this is not so basic), I've started going through Harrison Kinsley's "Neural Networks from Scratch in Python" which I think covers some of what you're saying.

@bosslogic

By all means, if you wanna learn the math, do so, but it's usually enough knowing the main properties of activation functions and the tradeoffs of having more or less layers.

Common ML frameworks will hide the math from you.