Creativity in machine learning teams
August 18, 2023
Some of my most notable moments in my career was when someone in the team pulls off something seemingly impossible over the weekend. On closer examination, the solution they came up with is just surprisingly and refreshingly creative compared to what you’ve been building the whole month.
I’ve seen people:
- repurpose well-known age old datasets for completely different problems
- reuse vision models on text problems
- train with just one example to prove that our model does not work at all
- use a 7 liner function that beats models in development
- bring down docker images to a fifth of their size
- add (or remove) just one line, thereby improving serving performance … and so much more.
Often, it just sets sparks flying all across the entire team, and you get a chain reaction of creative solutions from different people putting their heads together. And by the weekend you’ve made several leaps and breakthroughs and are much closer to delivering value.
These are not one off incidents- there are always specific people on the team who do this time and time again. Most AI teams today hire for deep knowledge in math or software packages. At the heart of it, machine learning is about problem solving. People who would bring fresh perspective into the team, by definition are helping you see and solve the problem in different ways.
So if you’re hiring, look for the quirky side project in the resume, ask them if they have built any websites or games on the side, see if they tell you a story about it.
If you’re leading the team, don’t tell people “that will never work”. Instead, ask them “how might we get that to work when we have 100 cuncurrent requests?“.
And who knows?
They might just surprise you on Monday.