I don't think I'll ever face such a dilemma

It's a question of risk aversity.

DeepMind is the safe option. It's prestigeous and a large company.

A startup is likely to fail. You should be prepared for that. In the unlikely case that the startup succeeds, I'd prefer the startup over DeepMind.

So it depends on how likely you think it is that the startup succeeds and how much you would "miss out" if it doesn't.

DeepMind, because then you're getting interviews much more easily for the next job.

Depends on what I'll be working on and who I'll be working with. I know plenty of brilliant people who work for big name companies that are complete asshats that I would never work with no matter how much money I was offered (unless it was 7 figures). If the startup is working on something I believe in and the people are likable, I'd choose the startup.

Remember the if in “IF the startup succeeds”. Also, If I knew stuff well enough that I could work at DeepMind, I would start my own ML startup. But first I would join DeepMind so that VCs will throw some cash at me.

DeepMind obviously.

> Assuming similar total compensation

This is a bad assumption, startups don't pay that well.

> if the startup succeeds then you get to be rich when the equity/options skyrocket.

This is the kind of crap they tell you to get away with giving you a lower salary.

You need to be pretty gullible to think that this is a close decision.

It is unlikely that you are going to personally shine at Deepmind. If you think that the startup is doing the right thing and that you can have an impact on it, then it is probably a better carrer choice.

Deepmind due to being the shining glimmer of success for others to aspire to where the startup could just be a massive waste of time

"Working for a no name startup so that your future HR looks at your CV and doubt you do anything much" vs "working for Deepmind so that your future HR says wow the moment they look at your work experience section". Yup this is a no brainer. Unless you have the ability to carry the startup, there is no reason to take a dangerous risk like that. Moreover, startups won't let you work on what you want easily because most of the time they need more people to work on the main products.