All of eugene_black's Comments + Replies

Any chance that Anthropic might expand the team to remote international collaboration in the future? I would apply but I am from Ukraine. Many great software companies successfully switched to remote work and covid crysis boosted this practice a lot. So just wondering.

2Andy Jones
It's not impossible, but it appears unlikely for the foreseeable future.  We do sponsor visas, but if that doesn't suit then I'd take a look at Cohere.ai, as they're one org I know of with a safety team who are fully-onboard with remote. 

Yeah, I think for general activities we can make a list of things that have a positive utilities for most cases. For example:

  1. Always care about your health and life. It is a base of everything. You can't do much if you are sick or dead.
  2. Don't do anything illegal. You can't do much if you are in prison.
  3. Keep good relationships with everybody if that does not take much effort. Social status and connections are useful for almost anything.
  4. Money and time is a universal currency. Try to maximize your hourly income, but leave enough space for other things from
... (read more)

Well, we definitely need a good definition of Morality then. And what is moral and non moral preferences. Looks like it converges to a discussion about terminology. Trying to understand what do you have in mind I can assume that an example of non moral preferences can be something like basic human needs. But when you choose to have this as a base doesn't that become your moral principles?

1TAG
That's not impossible ... we perhaps have too many candidates, not too few.. Is that a bad thing? If you don't discuss what you mean by "morality" you might end up believing that all preferences are moral preferences, just because you've never thought about what "moral" means .

But here is a problem: how would you calculate your utility if you have no moral system? You need at least more moral axioms.

1TAG
In the absence of morality, you maximise non moral preferences. There is no proof that all preferences are moral preferences. It doesn't follow from "all morality is preferences", even if that is true.

If you know believe that nothing is right do the following:

  1. Remember that nothing is 100% true so there is a chance that this is a false assumption. 
  2. Take all candidates for Morality that future you might follow.
  3. Make a weighted sum of normalized utility functions of every M. Take a somehow calculated (need to think how) probability of you choosing a specific M as a weight. 
  4. Normalize. 
  5. Zero utility function of nothing-is-rightness will not participate as you can't normalize constant zero. 
  6. You have an utility function now. Go and work. 
... (read more)
2tivelen
This is something I've thought about recently. Even if you cannot identify your goals, you still have to make choices. The difficult part is in determining the distribution of possible M. In the end, I think the best I've been able to do is to follow convergent instrumental goals that will maximize the probability of fulfilling any goal, regardless of the actual distribution of goals. It is necessary to let go of any ego as well, since you cannot care about yourself more than another person if you don't care about anything, now can you?