Gebru doesn't have anything good to say about AI, it's only downsides. She's biased being an activist and all.
Edit: to tie it back to the original criticism: what’s the maximum training cost we’re willing to accept for the model? How can we guarantee return greater than the increased training cost?