Are instruction-tuned, encoder-only transformers a thing again?

One user wonders whether the model approach is making a return, greeting the idea with a brief “cool.”