We start to use ML techniques to analyse the model outputs (eg to find tropical or extratropical cyclones). But the trend is also to replace bits of the model with ML trained on more complex process models so as to increase complexity while keeping the computational cost affordable.