That's downtown Denver next to the stadium and I-25. It's probably taken within the last few days as that's what the snow currently looks like. Don't know what that is.
>What does an AI/Machine Learning engineer do all day?
im not sure you could call me either of those but i spend most of my time aganizing over pdf parsing and how sam "salt-man" altman is salting the internet to frick with my training data
I am very disappointed by other answers in this thread. Makes you think
First, it really depends on company and how well processe are built. But anywhere, most of the time you work with data. You toss it around, clean, filter, reorder, restore, etc. Depending on the amount of data, and again processes, you use different tools and different level of automation. But data is what powers ML so most of the time you play around with it. Sometimes you even have to collect it (semiautomatically) from prod environment. Sometimes you have to look at data manually and decide if you can build a solution based on it. Sometimes you have to label the data, or find other people in company who can label it for you (usually it is done by low-paid QAs).
You second main task is actual experiments. You try different models and track everything so that you can rollback to your previous results. Hrnce, in some (bad) companies you first have to set up basic MLops environment.
Anyway, you try different models and modes of training. Sometimes there are tasks with clear objective function and you just fight to optimise it. But often you also have to decide what you need to optimize and explain to other departments why exactly, and ensure they agree with you.
Sometimes you have to combine a number of modules into one, sometimes you have to invent your own architecture. You also have to read a lot of scientific papers and decide which new approaches to try on your own data, as you cant invent everything by yourself. Sometimes you can use pre-learned model, but it is rarely the case.
You should possess many skills from different areas to be a valuable ML engineer.
I dont need any of those skills to engineer a dataset, the problem with most of the AI use cases I found that I know how to optimize is no one will pay for the services I could provide.
It's not about speed it's about the fact that no one else is doing this kind of thing because of a combination of laziness and also expecting others will do it eventually. the thing is most business people are either idiots with the ability to print free money to get out of any issue, or else are above my skillset and have no use for me. The former would never take a chance on a risky idea, and the latter needs people with actual learned skill. Most of the stuff you are describing though sounds like boring shit anyone could do. when it comes to employment, 99% of those money grubbing richtards I just described have egos the size of moons so they would never hire someone like me, they expect people who are actually smart to quit their coushy bleeding edge AItism job or pen testing group to come babysit a SEIM or install stable diffusion or write chatgpt API scripts for them.
they are the business equivelant of a used up prostitute who is ready to settle down asking for a 10 to come paypig for her.
>Most of the stuff you are describing though sounds like boring shit anyone could do
You can't escape boredom completely, that is true. Data processing is always very tedious for me, but is still necessary.
However, combining and inventing models requires you to be skilled. To start with, you HAVE to know state-of-the-art approaches for different tasks. Then, you have to know how to tune and combine them. And you have to have an intuition (drawn from your experience) of which approach would work and with would not.
I agree that most of the higher management are dumbasses. However, if the company is any good, there are should be people who care about product. You should stick to them. Communication is also a skill btw.
the last thing is the only thing you said that isn't moronic. look I get that technical skills are important but thats not the issue here, the best is a problem for later, good enough is fine. the real issue is communication. Most of these management morons are incapable of communication, to the point that they simply print more money whenever not enough people buy what they are making.
5 months ago
Anonymous
>the last thing is the only thing you said that isn't moronic
Thanks lol
Mostly wait for whatever they configured to finish training. Then find out they set one of the weights wrong, and then retrain all over again, this time for sure. This process repeats 5 more times before a good model is actually made. Except it was a model for simulating how pigeons shit or something equally inane.
Where is that?
Who is that from, what is that?
Where is it going?
That's downtown Denver next to the stadium and I-25. It's probably taken within the last few days as that's what the snow currently looks like. Don't know what that is.
Denver is a weird place. Around 2012 people were flocking to its airport in secret
You know who else is flocking here? Californians, Texans, and Californian-Texans.
You reap what you sow.
Your state is a gay and drug addict mecca
Source?
Because it was one of the last few airports that had an indoor place to smoke in.
blud u at palantir?
>What does an AI/Machine Learning engineer do all day?
im not sure you could call me either of those but i spend most of my time aganizing over pdf parsing and how sam "salt-man" altman is salting the internet to frick with my training data
get paid to proompt
He starts and stops training scripts every few minutes.
In between, he's sucking off his bosses dick.
CEO gives them prompts to put into SD
AI/ML monkey pastes and runs them, then picks out the best results for the CEO to coom to at the end of the day.
regex
I am very disappointed by other answers in this thread. Makes you think
First, it really depends on company and how well processe are built. But anywhere, most of the time you work with data. You toss it around, clean, filter, reorder, restore, etc. Depending on the amount of data, and again processes, you use different tools and different level of automation. But data is what powers ML so most of the time you play around with it. Sometimes you even have to collect it (semiautomatically) from prod environment. Sometimes you have to look at data manually and decide if you can build a solution based on it. Sometimes you have to label the data, or find other people in company who can label it for you (usually it is done by low-paid QAs).
You second main task is actual experiments. You try different models and track everything so that you can rollback to your previous results. Hrnce, in some (bad) companies you first have to set up basic MLops environment.
Anyway, you try different models and modes of training. Sometimes there are tasks with clear objective function and you just fight to optimise it. But often you also have to decide what you need to optimize and explain to other departments why exactly, and ensure they agree with you.
Sometimes you have to combine a number of modules into one, sometimes you have to invent your own architecture. You also have to read a lot of scientific papers and decide which new approaches to try on your own data, as you cant invent everything by yourself. Sometimes you can use pre-learned model, but it is rarely the case.
You should possess many skills from different areas to be a valuable ML engineer.
I dont need any of those skills to engineer a dataset, the problem with most of the AI use cases I found that I know how to optimize is no one will pay for the services I could provide.
If you know how to optimize better than other people, you could easily win Kaggle competitions couldn't you?
It's not about speed it's about the fact that no one else is doing this kind of thing because of a combination of laziness and also expecting others will do it eventually. the thing is most business people are either idiots with the ability to print free money to get out of any issue, or else are above my skillset and have no use for me. The former would never take a chance on a risky idea, and the latter needs people with actual learned skill. Most of the stuff you are describing though sounds like boring shit anyone could do. when it comes to employment, 99% of those money grubbing richtards I just described have egos the size of moons so they would never hire someone like me, they expect people who are actually smart to quit their coushy bleeding edge AItism job or pen testing group to come babysit a SEIM or install stable diffusion or write chatgpt API scripts for them.
they are the business equivelant of a used up prostitute who is ready to settle down asking for a 10 to come paypig for her.
>Most of the stuff you are describing though sounds like boring shit anyone could do
You can't escape boredom completely, that is true. Data processing is always very tedious for me, but is still necessary.
However, combining and inventing models requires you to be skilled. To start with, you HAVE to know state-of-the-art approaches for different tasks. Then, you have to know how to tune and combine them. And you have to have an intuition (drawn from your experience) of which approach would work and with would not.
I agree that most of the higher management are dumbasses. However, if the company is any good, there are should be people who care about product. You should stick to them. Communication is also a skill btw.
the last thing is the only thing you said that isn't moronic. look I get that technical skills are important but thats not the issue here, the best is a problem for later, good enough is fine. the real issue is communication. Most of these management morons are incapable of communication, to the point that they simply print more money whenever not enough people buy what they are making.
>the last thing is the only thing you said that isn't moronic
Thanks lol
proooooompt
Mostly wait for whatever they configured to finish training. Then find out they set one of the weights wrong, and then retrain all over again, this time for sure. This process repeats 5 more times before a good model is actually made. Except it was a model for simulating how pigeons shit or something equally inane.
Wait for a turn on the company compute cluster to run tests.