[D]Can a ML model be trained using other ML models as input data to try and develop novel architectures?

You'll have to excuse me if this is a daft question. I'm quite new to ML, but a thought occured to me about how a ML model might be able to assist in developing new undiscovered architectures.

So I thought I'd ask.

Tried to find some examples on Google but was returned very limited information.

The overall things I'm getting at is, could a ML model design itself a more efficient architecture?

Thanks.

architectures data machinelearning ml models