ML Engineering - end of life?
I started as Java Backend engineer after masters. Worked for 4 years and then transitioned to ML Engg about 3 years ago. Spent about 50-50 right now between backend and ml engineering. Increasingly thinking unless you are doing cutting edge work in places like openai (preferably with PhD in AI) probably better off getting back and sticking to conventional engineering roles either backend or full stack. Also seeing lots of ML Engineers from teams like Amazon Alexa for example who got laid off open to getting back to any kind of engineering role. Is the ML Engineering hype over for most in this role? Also when I refer to ML Engineering not referring to data science / data analyst roles but roles where you focus more on training and deploying deep learning models. TC: 330k (400k before stock crash)
I feel like ML engineer is very similar to data engineer
ML eng got from building feature stores and ML storage and serving into more specialized skills now like writing gpu based libraries for running models, specialized architectures for running models etc. Otherwise rest of those got automated by pipelines and infra available off the shelf or in aws/gcp.
I move seen both backend, front end, ML engineers, Data Scientists, Technical Recruiters, and more getting laid off That’s why it’s more important to level up to a senior level role or move into senior management to have a buffer from being the first round of cannon fodder
ML is becoming commoditized. You don’t need a bunch of ML engineers training models but you will always need people to figure out how to serve the models to customers.
It's just a backend problem. Generalists can easily handle it
ML engineers are still very much sought after in the market. I don't any engineers from ML teams got laid off during SNAP layoffs. This includes engineers who work on ML infra.
Unless you are working on GPU library and framework level.... You would think you are ML engr running some models but you are not real ML engr. 😂
Re: Amazon I guess that’s precisely my point. If you need to be working on ML infra stuff and frameworks it’s like any other backend engineering role where you have to build distributed systems (well maybe throw in some nuances about GPU, cuda, etc) .. over the past few years we saw a proliferation of roles which neither involved doing any ML research nor building cutting edge distributed systems but kind of doing mix of everything in the middle labeled as “ML engineering” .. I guess time might be up for those type of roles. I would be honest the only reason I switched over was those roles seemed to be paying more but I guess it was too good to last 😀the infra stuff is slowly getting more and more automated on the cloud. Companies like openai and google are happy sharing their state of art models publically so without much effort you can fine tune an existing model instead of wasting months building from scratch. And research is well proper research with publishing papers and stuff. Which means pretty much game over for 80% ml engineering roles
This is not the case for medical AI applications
Is it because it’s a niche domain and the healthcare domain knowledge is the differentiator?
Yes, too many in the market. Also, all bunch of ML scripts are readily available to train a model. SWEs have started doing it. So there's a threat for sure
Can say this as I am an applied scientist, the Alexa SDEs that got laid off had nothing to do with actual ML engg. The people who actually do ml engg have not been laid off. At best ml ops guys got laid off, who built stuff around the ml engg stuff. So no need to worry about supply increase from alexa side.
What is your definition of who really does ML engineering ?
I mean off you aren’t doing low level inference stuff or writing frameworks over tf or pytorch, then you are likely not doing ML engg. Too many “ml engg” in alexa are just SDEs who claim to be ML engg
Agree with assessment.. ML pipelines are well setup in AWS et al.. so domain knowledge is more important. So unless you are a Ph.D, you can be easily replaced
I learned just enough to be able to hold a discussion.