The rabbit hole can go deep, and probably isn't worth getting too fancy for single-digit hosts.  Fleets of thousands of spot instances benefit from the effort.  Like everything, dev-time vs runtime-complexity vs cost-efficiency is a tough balance.

When I was doing this often, I had different modes for "dev mode, which includes human-timeframe messing about" and "prod mode", which was only for monitored workloads.   In both cases, automating the "provision, spin up, and initial setup", as well as the "auto-shutdown if not measurably used for N minutes (60 was my default)" with a one-command script made my life much easier.

I've seen scripts (though I don't have links handy) that do this based on no active logins and no CPU load for X minutes as well.  On the other tack, I've seen a lot of one-off processes that trigger a shutdown when they complete (and write their output/logs to S3 or somewhere durable).  Often a Lambda is used for the control plane - it responds to signals and runs outside the actual host.

There's a big presumption there.  If he was a p-zombie to start with, he still has non-experience after the training.  We still have no experience-o-meter, or even a unit of measure that would apply.

For children without major brain abnormalities or injuries, who CAN talk about it, it's a pretty good assumption that they have experiences.  As you get more distant from your own structure, your assumptions about qualia should get more tentative.

Do you think that as each psychological continuations plays out, they'll remain identical to one another?

They'll differ from one another, and differ from their past singleton self.  Much like future-you differs from present-you.  Which one to privilege for what purposes, though, is completely arbitrary and not based on anything.  

Which psychological stream one-at-the-moment-of-brain-scan ends up in is a matter of chance.

I think this is a crux.  It's not a matter of chance, it's all of them.  They all have qualia.  They all have continuity back to the pre-upload self.  They have different continuity, but all of them have equally valid continuity.

Think of it like this: if one had one continuation in which one lived a perfect life, one would be guaranteed to live that perfect life. But if one had 10 copies in which one lived a perfect life, one does benefit at all. It's the average that matters.

Sure, just like if a parent has one child or 10 children, they have identical expectations. 

I think we're unlikely to converge here - our models seem too distant from each other to bridge.  Thanks for the post, though!

Reminder to all: thought experiments are limited in what you can learn.  Situations which are significantly out-of-domain for our evolved and trained experiences simply cannot be analyzed by our intuitions.  You can sometimes test a model to see if it remains useful in novel/fictional situations, but you really can't trust the results.

For real decisions and behaviors, details matter.  And thought experiments CANNOT provide the details, or they'd be just situations, not hypotheticals.

Once we identify an optimal SOA

This is quite difficult, even without switching costs or fear of change.   Definition of optimal is elusive, and most SOA have so many measurable and unmeasurable, correlated and uncorrelated factors to them that comparison is not directly possible.

Add to this the common moral beliefs (incorrect IMO, but still very common) of "inaction is less blameworthy than wrong action, and only slightly blameworthy compared to correct action", and there needs to be a pretty significant expected gain from switching in order to undertake it.

With that in mind, suppose you are asexual. Would you take a pill to make you not asexual?

I'm not asexual, but sex is less important to me than for most humans, as far as I can tell.  I know of no pills to shift in either direction that are actually effective and side-effect-free, and it's not meta-important to me enough to seek out change in either direction.  This does NOT mean that I judge it optimal, just that I think the risk and cost of adjusting myself to be higher than the value.

In fact, I suspect such pills would be very popular if they existed, and I would likely try them out if common, to find out if it's actually better in either direction.

You could make this argument about a LOT of things - for any trait or metric about yourself, why is this exact value the best one?  Wouldn't you like to raise or lower it?   In fact, most people DO attempt to change things about themselves.  It's just not actually as easy as taking a pill, so the cost of actually working toward a change is nonzero, and can't be handwaved away.

Wow, a lot of assumptions without much justification

Let's assume computationalism and the feasibility of brain scanning and mind upload. And let's suppose one is a person with a large compute budget. 

Already well into fiction.  

But one is not both. This means that when one is creating a copy one can treat it as a gamble: there's a 50% chance they find themselves in each of the continuations. 

There's a 100% chance that each of the continuations will find themselves to be ... themselves.  Do you have a mechanism to designate one as the "true" copy?  I don't.

What matters to one is then the average quality of one's continuations. 

Disagree, but I'm not sure that my preference (some aggregation function with declining marginal impact) is any more justifiable.  It's no less.

Before even a small fraction of one's life has played out, one's copy will bear no relation to oneself. To spend one's compute on this person, effectively a stranger, is just altruism. One would be better off donating the compute to ASI. 

Huh?   This supposes that one of them "really" is you, not the actual truth that they all are equal continuations of you.  Once they diverge, they're still closer to twin siblings to each other, and there is no fact that would elevate one as primary.  

This is a topic where macro and micro have a pretty big gap.  

If you're asking about measured large-group unemployment, you probably don't get very good causality from any given change, and there's no useful, simple model of the motivations and frictions of potential-employeers and potential-employees.  It's a very complicated matching market.

If you're asking about some specific reasons that an individual may be out of work or become out of work, you'll get a lot better result and some concrete reasons.   But everyone you talk to will say "that doesn't scale!".    

At its most useless modeling level, unemployment happens when some people don't want to (or aren't allowed to) accept the wage that someone can and will offer.  

I don't understand the question.  What intuition for not smoking are you talking about?  CDT prefers smoking.   Are you asking why EDT abstains from smoking?  I'm not the best defender, as I don't really think EDT is workable, but as I understand it EDT updates it's world state based on actions, meaning that it prefers the world where you don't have the lesion and don't WANT to smoke.