In 2012, python was already well-established in ML, though not as dominant as it is today. scikit-learn was already well-established and Theano was pretty popular. Most of the top entries on Kaggle were C++ or Python.

Julia only came on to my radar in scientific computing (not ML) in about 2015-2016 or so but while I tried it at the time, it was really not very stable and my view was that it was very immature compared to Python’s scientific ecosystem. Looking at the dates, v1.0 came out in 2018 and I remember going to a talk about it at my academic institution where someone showed off the progress and we had a play again in our research group but it still didn’t have many things we needed and the trade offs felt not great as we were heavy users of IPython and then when it came out Jupyter and while Ju stood for Julia the kernel development environment didn’t work so well because rerunning cells could often cause errors if you’d changed a type for e.g.

At the time we were part of the wave I suppose that was trying to convince people that open source Python was a better prospect than MATLAB which was where many people in physics/engineering were on interpreted languages. At least in my view, it wasn’t until much more recently that Julia became a workable alternative to those, regardless of the performance benefits (which were largely workable in Python and MATLAB anyway - and for us at least we were happy developing extension modules in C for the flexibility that the Python interface gave us over the top).

> Even 7-8 years ago people we're still having Python vs R debates.

They still have to this day.