Reprinted from: New Wisdom Yuan
With the continuous development of AI technology, more and more applications are available, and the salary of AI talents at home and abroad is also rising.
The Internet is also full of myths about creating wealth. For the freedom of wealth, not only programmers in various fields, but people from all walks of life have almost changed their careers and flocked to the AI industry with the wave.
But if you don’t like machine learning and can’t afford such a high salary, would you be willing to change jobs?
Most people’s choice may be to choose to compromise with life, after all, they give too much.
One Reddit user recently decided to quit machine learning for good.
Here’s what he says about himself, and how to find his next job that he likes.
I’m tired of machine learning
= = = = = = = = = = = = = = =
I’m one of the few people in the industry who does machine learning for a living but has no interest in it.
I’ve been developing models using classical and deep learning methods for 7-8 years, and many of these models have had a great impact in the companies I’ve worked for.
I think I’m pretty good at that, too, and I get paid pretty well.
But right now, there’s nothing in machine learning or deep learning that excites me.
I find it more enjoyable to solve problems in math textbooks. Actually, I want to do some kind of math career, but I don’t want to do machine learning all my life.
Before I went into machine learning “for the money,” I did a lot of work in satellite systems engineering. I also took many physics and EE courses (optics, quantum mechanics and Solid-state devices) during my master’s studies.
I’m thinking about working in quantum information, but I don’t have a PhD yet. Also, my computer science skills weren’t strong enough to transfer to cryptography.
So people, how do I get out of machine learning?
Don’t resign naked
= = = = = = = = = = = = = = = = = = = =
Once the post was posted, it immediately caused heated discussion among netizens.
“I know how boring it feels, but instead of leaving, you should change your job direction,” said Pruby. It can take months or years to figure out your next steps, so it’s best not to quit naked.
Shot_a_man_in_reno agrees, adding that you can learn another subject and apply machine learning techniques there. “I don’t want to work in general-purpose machine learning, like face monitoring or NLP,” he says. “People who graduate in computer science are doing pure stuff, so there’s a lot of competition.” A geneticist or anthropologist with dual expertise in machine learning might do something more interesting. ,
Another expressed why he thought the author hated machine learning.
MinLikelihood says, I love statistics/machine learning, but the reason I don’t want to have a long-term career in this field is this: I find processing and analyzing data boring. I like theory, and I like to explore Estimators, study sampling methods, and develop new optimization techniques instead of using them. Simple use is repetitive and will definitely get you bored after a few years.
‘Go ahead and be free,’ advises another.
According to Beexes, you should always only do what makes you happy. I have a friend who quit his job as a software engineer and started a restaurant. He should be the happiest man in the world now.
Some people think that the workplace is also besieged, you yearn for the place, for others also very boring.
EdAlexAguilar shared, “I got my MASTER’s degree, PhD and post-doc in quantum information, but now the research in this field doesn’t excite me anymore.” I have been doing intensive learning for the past year and I love this new field. It didn’t happen overnight. In fact, I knew during my PhD that I was likely to transition and leave, but wanted it to be as smooth as possible. My guess is, so do you. You don’t need a PhD, just extra energy to study. If you’re under financial pressure, start working on new goals in your spare time. It may take a few years, but you can manage to jump into different fields while you have a job.
There are also netizens from the perspective of society to analyze the problem of mathematics career.
Cookiemonster1020 says the lack of jobs in math/applied math is partly due to the hype behind big data. When I entered the academic market after graduating as a postdoctoral fellow at a prestigious university, I found that there were almost no openings in my applied mathematics field. Instead, people can get a job even if they’re just doing a compressed sensing application.
After listening to the comments, the creator of the post said that thanks to the netizens’ analysis, he has already started to read the online course on discrete differential geometry, and plans to explore the combination of differential geometry and machine learning in the future.
He says he would rather work in ML in a field such as applied physics or genetics than in a company such as banking, social media analytics or e-commerce.
As for the reason why I hate machine learning, I am tired of technical papers titled “X is all You need”. I have nothing against anyone who published the paper, and I am absolutely sure that the author is more qualified than I am, but I am very uncomfortable with the flashy title of the paper. Because I’ve never seen such grandiose titles in physics or mathematics, which is one reason I didn’t want to do a PhD in machine learning. I hate that style! !
And taking any online course from any platform will not make you a data scientist or ML researcher. Very few AI practitioners want to take the time and effort to learn the basic math behind machine learning algorithms. When I ask candidates in interviews to explain what PCA is, they only answer that PCA is a dimensionality reduction technique, with no mention of eigenvalues, eigenvectors or covariance matrices at all.
Artificial intelligence is a nuisance
= = = = = = = = = = = = = = = = = =
Not coincidentally, there is also a discussion on “why I don’t like AI” on Zhihu.
The monopoly of deep learning makes the term “algorithm engineer” controversial.
Some researchers have summarized the four deadly SINS of artificial intelligence research:
1. Technology update is too fast, and I will be out after three days without reading the paper
Learning is a good thing, but the rapid pace of iteration and the volume of papers make researchers frustrated, and one day’s dedicated learning may become obsolete tomorrow.
And it is easy to crash idea, or the SOTA model in hand has died before publication.
2. Excessive capital expenditure
Once the paradigm of mass pre-training was established, it was hard for the average researcher to get a piece of the cake.
It costs money to collect and download large-scale training data, it costs money to annotate data, and it costs money to stack the hardware required for training models.
Even if the data is big enough, the resources are strong enough, and the methods aren’t brilliant enough, your model will probably learn faster and perform better than anyone else.
3. Hard work on improving the model
The black box model of deep learning is a cliche, but it can be a mental torment for programmers.
Performance is not strong enough, convergence is not good enough, and the predicted results are not as good as expected, and the “alchemist” is crying in the face of models and parameters. Creating code can be fun, but tuning does not: performance improves, but how? No one knows!
For example, when you change the activation function from Relu to Selu without any performance improvement, you come up with a new SOTA model. Should you be happy or should you doubt yourself?
4. Data is everything
The current AI model is like a ruthless machine that can answer some of the questions in the exam paper if given enough data, but there is no progress in the understanding of knowledge and logic of the model.
Ng once proposed the 80/20 rule: algorithm engineers should spend 80% of their time collecting and cleaning data, with the remaining 20% working on models.
In this case, the feeling of being at the mercy of data is too unfriendly for researchers who specialise in modelling.
In fact, AI is still an application field with great potential. Deep learning has raised model performance to an unprecedented height. Voice assistant, intelligent recommendation, face recognition and so on are making life more convenient.
But everything has its flaws. Do you think AI is “annoying”?
References:
www.reddit.com/r/MachineLe…
Finally, I welcome you to follow my wechat official account: Duibainotes, which tracks the forefront of machine learning such as NLP, recommendation system and comparative learning. I also share my entrepreneurial experience and life perception on a daily basis. Students who want to further communicate can also add my wechat account to discuss technical problems with me, thank you!