[D][R] How do researchers (Masters, PhD) implement complex models? Are they gods?

I'm doing my theisis right now. I have good grasp of the high-level details on most ML models (RNN, CNN, LSTM, Transformers, GPT, CNN, GANs, LDMs, VAEs, Autoencoder and much more). Of course by no means i'm an expert, but I'm able to learn what I need.

But when it comes to actually use them, and implement them in code, and train them, this becomes hell. For the simpler models, its fine, but for the more complex once, there are …

autoencoder cnn course expert gans good gpt learn lstm machinelearning masters ml models of course phd researchers rnn transformers