I agree with you from financials perspective. I see my peers having joined SWE posts in faang straight after bachelor's have been promoted to SDE2 and equivalent and live a very comfortable life.

I somewhat agree, my thesis also on very well defined properties of robustness in ML models but no explanation for larger models. We give theory for toy 2-3 layer networks with activations on hidden layers. However, recent progress in neural tangent kernels to explain DL is real.

I understand the faang argument. I also had a soft offer to join Amazon as a Applied Scientist I with just MS upon my graduation. However, that ship sailed due to Alexa hiring freeze. My prospective PhD program is completely rooted in academia. Maybe I can do some internship but my program doesn't have a industry collaboration.