Do large companies write their own neurons or take pre-trained ones?

Question

Neural networks

At the moment I am studying Data Scientista, and I am interested in the following - how are neural networks developed in any companies (large or small), namely, is the architecture developed from scratch or are ready-made models taken like ResNet, YOLO or the same BERT?

Answer 1

of course, those pre-trained
for tasks are already being corrected
- in the form of an exception and a new word, it is expensive and the output is usually zilch (well, except for Watson and what others have for the inside)

Answer 2

It happens this way and that. Only now, using exclusively ready-made models, you can consider yourself a Data Science specialist only in your imagination. At best - Data Analyst. Approximately in the same way that no one will call a person who uses programs written by someone a Programmer.