Sealed predictions thread

post by Zach Stein-Perlman · 2022-05-07T18:00:04.705Z · LW · GW · 6 comments

Want to publicly register predictions or other statements without yet revealing what they say [LW · GW]? Do so here. The advantage of commenting here (in addition to, say, your Shortform) is that I'll record all comments here so they can't be deleted or edited later, and I may publicly shame you if you don't reveal your statement when you say you will.

So, comment with

And, of course, save the statement. (If you use hashing and don't save your exact statement, you of course won't be able to prove it, so not only will your registration be useless, but also it will seem as if you post-facto decided to hide your statement.) If there are some circumstances in which you would not reveal the statement--e.g., if it could be infohazardous--strongly consider

Otherwise, it will look bad if the date arrives and you don't reveal your statement.

To retract or edit a statement, send me a private message within 24 hours of commenting; after that, it cannot be changed in my record (although you can of course comment to say that you no longer endorse it, or to add hashes with new statements).

Fancy methods may exist if you want to provably commit to sharing your statement.

It's your responsibility not to share infohazards. This includes ensuring that any hashes you share can't be reversed, even if computing power advances more than you expect in the time before you reveal your statement, and ensuring that you don't commit to sharing something that, it might turn out, you should not share.

Edit: it seems there are multiple ways to handle verification, but for now, here's the simplest, I think: a sheet with most of the relevant information from comments. Thus information cannot be edited or deleted (without my complicity). And I guess now that there's a sheet, you can PM me rather than comment here to be added if you prefer.


Comments sorted by top scores.

comment by Yoav Ravid · 2022-05-07T19:31:59.496Z · LW(p) · GW(p)

A fun, simple prediction just to participate. Will reveal on the 1st of June, 2022.

Hash: SHA-512


Replies from: Yoav Ravid, Yoav Ravid
comment by Yoav Ravid · 2022-06-01T12:55:21.952Z · LW(p) · GW(p)

My prediction was:

There will be less than 15 messages time-locked messages in this thread by the time this prediction is revealed (80% confidence)

And seems like that typo ("messages time-locked messages") was there, but I can't fix it because it would change the hash, lol.

comment by Yoav Ravid · 2022-05-07T19:37:31.891Z · LW(p) · GW(p)

Even though I made a simple prediction of no consequence and said so, I still have a nagging worry that for some reason I wouldn't be able to reveal it when the time comes (I lose the message, I'm unavailable, I made a mistake while copy pasting...) and it will look bad. I thought that was interesting to note. 

Replies from: quintin-pope, Zach Stein-Perlman
comment by Quintin Pope (quintin-pope) · 2022-05-08T03:22:16.862Z · LW(p) · GW(p)

I'd suggest writing the message as a plaintext file, then computing the hash of that file. That way, there's no possible issue with copy/pasting. You still have to track the file, but that's not too hard.

comment by Zach Stein-Perlman · 2022-05-07T19:55:40.600Z · LW(p) · GW(p)

Yes, this seems interesting and important. Mentioning the scope of the statement can help, at least -- in this case, you won't really look bad if you fail to reveal it because you mentioned in advance that it's trivial. But making sure you'll be able to reveal it later is quite important if registering statements for nontrivial purposes. (But this seems doable without much effort.)

comment by Zach Stein-Perlman · 2022-05-07T18:00:40.750Z · LW(p) · GW(p)

Will reveal by: 1 January 2030

Hash: SHA-512: e18e929b785bd66431d2cf23c16690dadbab6b0a695ad2b4fa1bf33ab89e093866a2a6f04afde0b5ca15ccc8048a043384bfe952e6769f7a85ad56d9dda23198

This statement is a policy of mine -- related to this post, in fact. I don't want to share it now, but I might later want to prove I planned it in advance, and I am essentially certain I will not mind sharing it in several years.

Posting hashes of personal policies, to be shared if it turns out they are relevant, seems like a use case of comparable value to registering predictions. In such cases, I'm not sure under what conditions it's best to add a reveal-by date -- I did so here because I'm sure I won't mind sharing in several years, but I don't want to set such a norm without more consideration.