Should I pick the newest model?

People like picking sides. And if you survey the NLU space, you'll find strong opinions. Deep learning is the future! Rules are explainable! And every combination in between. Talking about models is difficult because strong opinions tend to be grounded in truth.

  • there really are incredible differences in capability between model types
  • some models really are "old" while others really are "upcoming"
  • the experience of training and running models can vary incredibly

But rather than anointing some technology as the "the winner," it's best to take a pragmatic stance:

  1. The best model is the one that works for you, and
  2. The model that works for you depends on your situation

This isn't a suggestion not to form an opinion. On the contrary, grounding your opinions in the actual problems you need solved will result in stronger views, better defended. Go with what works best for you, regardless of how new or old it is.