The real difference between academia vs industry (by an ex-Uber researcher)
I recently talked to my friend Jason, who did research at Uber AI Labs, and later, he cofounded a startup doing ML for wind farms. If you are curious about the difference between the research world and the industry, keep reading. By now, you probably saw a lot of news about LLMs in research, but what exactly are the use cases for the industry? Here is a free workshop on "How to Build an LLM App on Inventory, Product & Reviews Data" on Oct 24th. Register here. (You'll still get the recording if you don't show up, as long as you register).
Here is my friend's take on ML in the academia vs in the industry:
As an ML researcher, your products are paper. Essentially, you're selling an idea, you're selling a story, you're selling a narrative. You hope it sticks in the market when people understand it. You want other researchers to adopt it, and it changes their research trajectory. And you get an exchange -- citations. The metric is popularity.
In the industry, the more product-based world you're selling products, and your products have to work. They have to bring value to someone, to some customer. The metric is revenue, not popularity. There are many successful companies out there that you've never heard of because popularity doesn't matter to them! They make some widgets and they sell them by the millions. They make tons of money and nobody knows about it. It's a different world.
Researchers sell ideas and papers, while ML practitioners in the industry sell products.
It's not that one track is better than the other; it depends on what you want in the current stage of your career. If you are interested in research and want to do a PhD, my friend also shared his experience as a PhD at Cornell University.
He spent six years getting his PhD and it's full of learning. However, only 25% of that learning has to do with the topic that you're working on. Besides the technical skills, you learn how to communicate, how to write, how to speak. But more importantly, you learn how to look at the world to figure out two things:
What questions could you ask that are interesting?
What strategies do you use to answer them well and answer them convincingly?
Some people learn and some people don't. It is to learn to see research results completely disconnected from your own ego. Maybe it's helpful to discuss the difference between science and engineering.
In engineering in the real world, the question you're mostly asking is, can we build something that works as well? And then you answer it by building something that works that well. And if you don't, then you feel like you've failed; you tried to build it and your latency is not good enough, or the memory is too much, or there are so many bugs in the system.
As a scientist, you try to ask questions, answer questions, and then you try to disconnect your ego from the answer so that you try to answer the question. Whatever the data is, you look at it and you say, "Oh, that's interesting; this works; this doesn't work."
Some people still connect the ego to the answer when they really want their research to work, and they're not as effective as scientists. They don't have the ability to look at the data and see the results as it is, and they might get frustrated when something doesn't work the way they want it to work.
Scientists want to learn; engineers want things to work.
Today, many PhDs have also become founders to build things, especially in the Gen AI area. There are companies acquiring researchers from labs of universities to get access to talent and tech. I don't think a lot of PhDs today still stay in academia, and there are a lot of investors trying to network with researchers in labs and the journey from a researcher to a builder is getting easier and easier.
One thing a lot of researchers are learning when building companies is to focus on how to make things work, even if the methodology is boring.
What's your experience with academia and the industry? Reply and let me know what you think! (Don't forget to register the LLMs for inventory and review data here!)
Until next time,
(I'll write more about different ML/DS roles, and emerging AI careers, stay tuned!)