Timeless Decision Theory Paper Released | Hacker News Reader

Timeless Decision Theory Paper Released

singinst.org

46 pointsXelaz13 comments

The linked paper, TDT-v01o.pdf, is bit-identical to a draft I downloaded in November. Which is odd since I thought I remembered a few drafty features in it like incomplete references to be filled in later, and it ends with "This manuscript was cut off here" and a one-entry bibliography.

Did they link to the wrong version?

You do realize that the linked article dates from November the 12th, do you? It linked to the most recent version which was then available.

I didn't notice -- thanks. I only skimmed the announcement and got the impression of a polished version ready to submit to some fancy journal.

It's interesting as is, anyway, judging by the first half. I'll be getting back to it.

Did they link to the wrong version?

It certainly looks like it. Maybe if Eliezer sees this he can post a link to the actual finished copy?

FWIW, though, whether finished or not there's a lot of really interesting stuff in there, definitely worth a read, even if you know/care nothing about/for decision theory.

Indeed! This paper seeems to have been so highly anticipated, it was released by those anticipating it before it was finished.

The title of the paper is a bit misleading, and rhetorically clever if intentional. When I read this headline the first time, I parsed it as "Timeless (Decision Theory Paper) Released", so that "timeless" was an adjective describing the (non-specific) "decision theory paper", which had apparently just been released. :-)

I don't think it's intentional - TDT is so named with good reason. I'm not sure I can explain why at almost 5AM, but it has to do with sort of treating one-shot games as iterated games, or something roughly similar in spirit to that. One way to think about it is to imagine that you may be a simulation of yourself being used to predict the real you's behavior, and therefore to adopt the behavior that you would want the simulator/predictor to predict from you. Now, if you're not a simulation of yourself, you still need to adopt that decision theory so that when you are simulated, the simulations behave that way. At least, that's how it solves Newcomb's Problem - that's quite possibly not a good overall summary of TDT.

I know I'm not supposed to snark. But teach yourself LaTeX, Mr. Yudkowsky. It'll be a gift to us all.

Okay, I'll bite.

What do I need to read before reading this to get a grasp of what's going on?

A good start is the LW Decision Theory sequence: http://wiki.lesswrong.com/wiki/Decision_theory_%28sequence%2...

See 'Decision theory' section of this reading list: http://lesswrong.com/lw/2ua/recommended_reading_for_friendly...

Deleted comment

If you read the provided background links, and maybe read up on Newcombe's Problem, you should be okay...I think.

If you want to get a better grasp, the standard LW reply is "read the sequences": http://wiki.lesswrong.com/wiki/Sequences

Reply on news.ycombinator.com