Top neural net engineers are getting paid $5-20M annual compensation packages | Hacker News Reader

Top neural net engineers are getting paid $5-20M annual compensation packages

twitter.com

39 pointsbarry-cotter17 comments

When I see neural net technical articles on hacker news there is not so much technical discussion anymore. Maybe all those hackers are busy doing their $10M jobs, or maybe they went to discord or to mastodon groups. Only the dumb ones like me are left. Just kidding I'm not dumb please headhunt me for $5M+

It's just what happens in these types of forums when a topic becomes popular. Mostly laypersons (typical HN user wrt ML) read the articles and comments, they vote based on the layperson's opinions, technical people's voices become lost in the mix or get downvoted because the layperson doesn't understand what they're talking about.

So what’s the next HN?

Twitter.

Good joke

> Only the dumb ones like me are left

It is worse on Twitter.

Hard to see this as anything other than a hang over from ZIRP. Yes, it makes total sense if you've got a billion dollars to throw at an AI start up that you can start just throwing absurd sums of money at engineers. Especially if you've convinced yourself (and your investors) that one of these companies is basically going to own the singularity.

What's worrying though, is this looks like a sort of wave of feedback loops. We go through these hype cycles, we had the uber economy, we had self-driving eating the world, we had crypto replacing the USD. And each hype cycle has gotten larger, and closer together. Each hype cycle more grand than the rest - this hype cycle is basically promising the end of the world. What happens when these technologies do deliver something - but not the singularity. What is the next hype cycle? Or do we finally face a real, massive bust, where investors have gotten so burned at this point that they really do start walking away from these stories.

So I have a CS degree from a state school, 30 years old, making 170k/yr at a lab doing scientific computing, should I just drop everything and try to get into a good school for deep learning like MIT?

Honestly it's possible that your scientific computing skills and experience would be more valuable. For example imagine doing a whole new deep learning degree program and they are just teaching you logarithms and some matrix calculus derivatives and how to use 'pip' that you already know right now.

You don't go to MIT for the education. There is absolutely nothing you will learn at MIT that you can't learn on your own or at any other college.

You go to MIT because MIT is a social filter that gives you access to world class opportunities that are almost impossible to get otherwise. Anyone who has an opportunity to go to MIT should do whatever they can to try to get in.

> You don't go to MIT for the education. There is absolutely nothing you will learn at MIT that you can't learn on your own or at any other college.

This is true if you limit it to the typical (non-research) undergraduate experience. But there are experts from top universities whose work is not accessible outside of academia. More practically, the university atmosphere also acts as a forcing function due to the requirements and competition. Add to that the fast feedback loop provided by some of the top minds (both teachers and peers). The chance that a self-taught engineer is as effective in their chosen field as an MIT graduate with even average grades is exceedingly low.

> You go to MIT because MIT is a social filter

Many go to MIT because it is (or at least they believe it is) the place to further their learning beyond everything else available. You're implying the social filter and network is the only benefit, which is simply not true. Even controlling for what it takes to be accepted in the first place, do you honestly think that MIT graduates do not gain anything from their time attending the institution other than the bona fides?

Depending on YOE you still have a lot of room to grow in terms of compensation (see https://www.levels.fyi/). You don't even have to go to FAANG. The hard part IMO is getting an interview, which is solved by getting a referral. Get someone in your network to vouch for you and secure an interview.

I give no advice for moving into NN work, there's a dearth of those engineers on the market, and a lot of jobs requiring N years of prior experience with a graduate degree (good luck hiring!).

Sure or do some independent research on DNNs and publish it. Since you are already in a lab you probably have some data there and any DNN work will likely be interesting. Use that to get into the field.

For the people working in Generative AI/LLMs, it is the best market ever. Even the end of 2021 wasn't anywhere near as crazy as this is. No surprise at all at these comp packages.

The top anything in any field will get this. Top actors are getting 10-20 mil per movie. Top interior design artists are earning 10-20mil a year.

This says very little about the job and more about cost of the top. Not sure what the point is if this tweet. Be the top of your game and make a tonne of money.

So this is the future? This is where the money is?

For now, up until the models can effectively maintain themselves, and then the money goes to whichever AI companies said engineers helped entrench unless there's a new taxation scheme on AI based on displaced labor.

So, if you do it and clear 5m a year for 3-5 years, at least you know you can retire peacefully in the new hellscape.

Reply on news.ycombinator.com