3 ms·

I think this depends on what you view as “getting screwed.” If the under the hood line engineers aren’t getting wild compensation for products they built gener

I think this depends on what you view as “getting screwed.” If the under the hood line engineers aren’t getting wild compensation for products they built generating wild compensation for the engineers/researchers working on the end product, yes that is a fair point. If it’s these engineers not getting the recognition of Chief AI scientist and so on, that’s misguided I think. There is nothing worse than working under a technical SME who doesn’t know leadership, and also “product,” broadly defined. I’ve worked on both sides of the issue and org leadership is a defined skill set the same way optimized code is. Unless the backend engineers display that leadership ability, their closeness to the internals of the technology doesn’t justify them running how the technology gets deployed. It is the same as if you found a startup. During the ideation sessions with whatever VC buys into you, there is 5% discussion of tech, or better yet - checkbox the product mvp works. 95% who are you going to sell this to and have you spoken to them? Different skills. When tech is a business, the tech knowledge starts to matter less, or is only a portion of the picture. The individuals that grasp that do well, SWEs who ignore this don’t IMO.

dauertewigkeit 3y ago
I don't think a CUDA engineer, or a researcher working on optimization or distributed computing should get the same recognition as a Chief of anything. But dare I say that writing optimized code to run on clusters is just as hard as coming up with a new deep learning architecture? Actually having done both, it is harder, at least for me. The authors of "Attention is all you need" got all the recognition. The people who worked on scaling deep learning models to huge clusters got no recognition. But both were just as instrumental to the ultimate success of LLMs.
- dogman144 3y ago
  Recognition and leadership roles isn’t based on hardness of the completed task. Never has been. A good leader will make sure the credit is spread out though, yes. Why: - Really hard or really easy things can generate the same revenue - Chief of X can screw up the leadership with hard or easy products if they can’t lead and explain to external audiences - external audiences deal with a million things to be an audience to. It’s wishing on a dream that that audience will dig into “well who actually built this” vs moving on to w/e other thing is trying to capture their attention. Put that together, recognition comes from glueing things and people together successfully, explaining how it works to outside world/clients, and steering the product away from an optimized elegant solution to w/e people will actually use. Ever since tech stopped being academic/open source and got revenue linked to it, “thems the rules.” Speaking of SWE survival guides, engineers often seem to ignore, hate, or argue this topic always, and meanwhile the smaller set of engineers who get on board with this (ugly, yes) reality tend to do well.

dauertewigkeit 3y ago

I don't think a CUDA engineer, or a researcher working on optimization or distributed computing should get the same recognition as a Chief of anything. But dare I say that writing optimized code to run on clusters is just as hard as coming up with a new deep learning architecture? Actually having done both, it is harder, at least for me. The authors of "Attention is all you need" got all the recognition. The people who worked on scaling deep learning models to huge clusters got no recognition. But both were just as instrumental to the ultimate success of LLMs.

dogman144 3y ago
Recognition and leadership roles isn’t based on hardness of the completed task. Never has been. A good leader will make sure the credit is spread out though, yes. Why: - Really hard or really easy things can generate the same revenue - Chief of X can screw up the leadership with hard or easy products if they can’t lead and explain to external audiences - external audiences deal with a million things to be an audience to. It’s wishing on a dream that that audience will dig into “well who actually built this” vs moving on to w/e other thing is trying to capture their attention. Put that together, recognition comes from glueing things and people together successfully, explaining how it works to outside world/clients, and steering the product away from an optimized elegant solution to w/e people will actually use. Ever since tech stopped being academic/open source and got revenue linked to it, “thems the rules.” Speaking of SWE survival guides, engineers often seem to ignore, hate, or argue this topic always, and meanwhile the smaller set of engineers who get on board with this (ugly, yes) reality tend to do well.