I have had a depressing day as we did a round of interviews for our machine learning positions
It was incredibly sad to talk to people who spent lakhs on ML courses to learn algos..and finally be shown AutoML
The second thing that shocks them is the pricing of Cloud Vision for face recognition - $1 for 1000 images. Or GPT3 - about 30k INR per month. About an intern's salary. @DataRobot costs less than a fresher data scientist for a year.
"Even with all the resources of a great machine learning expert, most of the gains come from great features, not great algorithms"
P.S. AutoML does features too :(
One answer: business lifecycle.
It is not important what algo u choose to go live. What happens after that ?
Uber's ML projects under the Michaelangelo umbrella are the formalized philosophy around this.
My prediction (no pun intended) is - data scientists will become closer to product managers.
They will own the onus of "conversion" & the "algorithms" ?
Justify ROI on cost of training ? P.S. GPT3 costs 5 mil USD to train.
And their salary scales with profit/conversion upside
Are you an ML scientist who can think business?
Build a career...stop paying for overpriced ML courses.
You may disagree with me. Sure algorithmic work has value - only in very specific orgs.
Not because those scientists are not smart - You guys are brilliant
But Google has made it useless.
Thought experiment - if u had a company, would you have opensourced Tensorflow or Pytorch?
Recommendation engine? AWS Recommend.
Image recognition - Cloud Vision or Rekognition.
NLP? GPT3
Voice ? Polly/Alexa.
Datarobot, AutoML, H20 Driverless AI cleans up the rest.
U need to start thinking business.
I have a new term for this - I call it "Product-Data Fit"
Here's a *specific* recommendation for you - read up everything on Uber's Michaleangelo components.