Oh. Yeah. I work as a systems guy on AI related tasks, but I have a strong mathematical background (my degrees are all in math), so I understand the AI stuff well. I just find the management of clusters and reliability engineering to be far more interesting than getting a computer to hallucinate nonsense. Anyway, the systems people are always dunking on the AI people for not knowing the basics of software like using ssh, setting up a firewall, or using version control software. We say things like, “yeah, but remember that the AI guys are the users, so we have to make it idiot proof” and “what’s the difference between malware and a neural network? Not much, but one only runs on Nvidia”
Previously, I was the platform engineer for a self driving car project owned by a very large vehicle OEM. I would never get in a Tesla using “full self driving” and neither would any of my colleagues then or now. By the way, that self driving project collapsed because a very capable car manufacturer that produces more vehicles in a day than Tesla produces in a year realized it would never work at the consumer level and we had the benefit of a half a million dollars worth of military grade localization equipment, while Tesla is trying to get by on webcams and magic.
Hell, you can pull up countless, peer reviewed papers in the AI field that don’t have error bars or any statistical hypothesis testing at all, authored by labs at MIT or Facebook. It’s terrifyingly bad.
tl;dr AI people are generally the least competent people I’ve met in my field-- which is AI.
also, I think most of their competent hackers leave. I work with a three Iranian systems dudes (not in Iran, despite the suggestion of another user here). If your choice is between working for the IRGC (becoming the target of western sanctions and security agencies) or a $250k/year job at Google, the choice is usually pretty easy.
Oh. Yeah. I work as a systems guy on AI related tasks, but I have a strong mathematical background (my degrees are all in math), so I understand the AI stuff well. I just find the management of clusters and reliability engineering to be far more interesting than getting a computer to hallucinate nonsense. Anyway, the systems people are always dunking on the AI people for not knowing the basics of software like using ssh, setting up a firewall, or using version control software. We say things like, “yeah, but remember that the AI guys are the users, so we have to make it idiot proof” and “what’s the difference between malware and a neural network? Not much, but one only runs on Nvidia”
Previously, I was the platform engineer for a self driving car project owned by a very large vehicle OEM. I would never get in a Tesla using “full self driving” and neither would any of my colleagues then or now. By the way, that self driving project collapsed because a very capable car manufacturer that produces more vehicles in a day than Tesla produces in a year realized it would never work at the consumer level and we had the benefit of a half a million dollars worth of military grade localization equipment, while Tesla is trying to get by on webcams and magic.
Hell, you can pull up countless, peer reviewed papers in the AI field that don’t have error bars or any statistical hypothesis testing at all, authored by labs at MIT or Facebook. It’s terrifyingly bad.
tl;dr AI people are generally the least competent people I’ve met in my field-- which is AI.
also, I think most of their competent hackers leave. I work with a three Iranian systems dudes (not in Iran, despite the suggestion of another user here). If your choice is between working for the IRGC (becoming the target of western sanctions and security agencies) or a $250k/year job at Google, the choice is usually pretty easy.