I am open to being an outside advisor / buddy / contact etc to individuals within this and/or with the project as a whole.
Me too!
Partially, I'm afraid that if this doesn't go well, our community will lose a cohort of promising people.
I really don't know what you mean by "lose" here (and I'm worried that others will have varying interpretations as well). Do you mean they'll become less promising? Not promising? Leave the community? Go crazy? Die?
Anyway, this seems sensible, but I still want to nudge you and everyone else in the direction of sharing more explicit models of what you think could actually go wrong.
Sorry, I was imagining a scenario where a person has an unpleasant experience and then leaves the community because for the last several months all their close contacts in the community were in the context of an unpleasant living situation. That's bad for the person, and unfortunate for the community as well.
I've come around somewhat to the outside buddy idea below; I dunno about the buddies knowing each other. That seems to introduce a whole new layer of difficulty, unless you're just talking about, like, an email list.
Cool. Yes, a mailing list sounds even better than the low-tech solution I had in mind, which was "every buddy learns 80% of the names of the other buddies through the grapevine, and they happen to be one or two hops away on the social network".
Sure, but what I'd like to know is why Nisan thinks that difference is important in this case.
I'm not proposing a house policy here. I'm suggesting that a Dragon would do well to have regular followups with someone outside the house, and I'm proposing that some members of the wider community offer to be those someones.
In the past I've had regular video calls with a couple people who were doing long-term experiments with their lifestyle; I think it was helpful. I believe such an arrangement was part of the Leverage polyphasic sleep experiment.
Jacob is right: There's a difference between a friend one can reach out to if one needs to, and a friend one is scheduled to talk to once a week. Personally, I struggle to keep up with friends without scheduled meetings, and it sounds like the Dragon Army will be very busy.
Also, there is a difference between reaching out to a friend when things have gone very wrong and one needs to get out; and bringing up a less drastic problem during a weekly check-in. In the first case, you need a couch to crash on and maybe a lawyer. In the second case, you need someone who will listen to you and bring an outside perspective, and maybe refer you to other resources.
Partially, I'm afraid that if this doesn't go well, our community will lose a cohort of promising people. It would be a shame if that happened because we failed to pay attention to how they were doing.
But also, if the experiment goes very well, this arrangement would be a means by which the wider community can learn from what went right.
In response to
Dragon Army: Theory & Charter (30min read)
Are there people external to the project who are going to keep an eye on this? I think it would be sensible for each participant to have a buddy outside the house who checks in with them regularly. And for each buddy to know who the other buddies are.
In response to
Superintelligence: The Idea That Eats Smart People
It's a curious refutation. The author says that the people who are concerned about superintelligence are very smart, the top of the industry. They give many counterarguments, most of which can be easily refuted. It's as if they wanted to make people more concerned about superintelligence, while claiming to argue the opposite. And then they link directly to MIRI's donation page.
In response to
CFAR's new mission statement (on our website)
Maybe you'll cover this in a future post, but I'm curious about the outcomes of CFAR's past AI-specific workshops, especially "CFAR for ML Researchers" and the "Workshop on AI Safety Strategy".
In response to
The correct response to uncertainty is *not* half-speed
Imagine an undirected graph where each node has a left and right neighbor (so it's an infinitely long chain). You are on a node in this graph, and somewhere to the left or right of you is a hotel (50/50 chance to be in either direction). You don't know how far -- k steps for an arbitrarily large k that an adversary picks after learning how you will look for a hotel.
The solution that takes 1 step left, 2 steps right, 3 steps left, etc. will find the hotel in O(k^2) steps. Is it possible to do better?
I can get O(k).
In response to
Open Thread, Jul. 6 - Jul. 12, 2015
View more: Next
Can I get contact info from you? I already have Malcolm's; if there's an email address you can use to send a message to TK17Studios at gmail dot com, I can then offer that address to anyone without an obvious check-in.
Sent.