Scared tech workers are scrambling to reinvent themselves as AI experts (opens in new tab)

https://web.archive.org/web/20190428005007/https://min.io/

if the work you're doing is work that you're genuinely interested in, you have a huge immediate advantage, and

Running into where crowd is running out, and running out of where the crowd is running in usually (but not always) gives you a large advantage.

Actually developing models is a much smaller field (in terms of jobs) and the overlap with software engineering is small. I’m sure you can make a lot of money in it, but you’re better off coming at it from data science.

And by far the most deployed technologies, even for postdocs, are going to be basically off-the-shelf or somewhat fine-tuned existing models via API or tooling being used for text and/or image generation for specific applications. In other words, things that in no way require the degree. That isn't to say that having real machine learning specialists isn't desirable, but there is plenty of room for people who just know how to apply the tools rather than invent new ones.

Actually there are plenty of "academic" papers coming out from people with advanced degrees that are mainly just applying GPT-4 to write code in some DSL to solve some category of problem. That stuff doesn't even require basic understanding of neural network architectures.

The latest models are so powerful it seems to be having a democratizing effect rather than increasing the class divide. From my perspective.

Anyone taking out loans for a PhD in the UK is a fool. Not only are our PhDs generally less well regarded than those from the EU and US, but there is also an abundance of grant funding available. If you can't get paid as part of a grant (and hence have fees taken care of) for a PhD here, then either the field of study has little economic value, or you are not considered good enough.

Even at fairly mature companies, fully utilizing a top tier ML expert may require the equivalent of 6-10 SWEs.

Anyway, while no one knows what the future holds, right now it doesn't seem like many companies are looking to actually take on AI as a core capability, they're looking to leverage open source tools. I'm at a large company right now that's betting pretty heavily on AI and we're thinking of bringing on 1 maybe 2 ML specialists over the next year. My friends at other large tech firms report similar plans.

Oh, and you need to be able to do all the above _quickly_.

What's nice about it versus SWE is that you can totally ship hot garbage, doesn't have to be optimal at all, just "fast enough" and cheap enough.

Modern ML has been totally different for me. It was janky before the craze, it is still janky as heck, and the ML community has its own unique blend of manic, reckless myopia and scammers. But the deeper I go, the more mind boggling it gets.