Neel Nanda comments on Deep learning curriculum for large language model alignment

[Neel Nanda]

That seems like a pretty reasonable breakdown (though note that 2 inherently needs to come after 1)