I believe Eliezer has the best overall understanding of the issues related to alignment.
Also, I'm confident there is a solution.
This will only resolve if Eliezer looks at my algorithm.
If he does, this will only resolve yes if, after review, his new p(doom) is < 0.1. (10%)
It will only resolve yes with Eliezer's consent.
It will resolve no otherwise.
2 min general overview by request.
Update: Due to integrity and logistic concerns in the comments, the above prediction will resolve N/A on Jan 1st 2030.
Here's an argument if anyone would like to engage with the claims I'm asserting.
https://manifold.markets/Krantz/which-proposition-will-be-denied?r=S3JhbnR6
I don't believe I am able to do that. It would need to be an admin correct?
Overall, I am open to whatever method incentivises users to produce an accurate prediction.
Resolving to NA seems like it would produce many comments criticizing how it would be a complete waste of time (assuming they had a high confidence that it would default to na).
With partial, people could at least bet it down to 2% and profit from the wagers I've already placed.
Also, this isn't Rob Miles, correct?
If you are, please just provide me a secure place to send you a document (after somehow demonstrating that you are).
@Krantz One problem with resolve-to-market is that the incentives are weird - e.g. someone could bet it up not because they think the true probability should be higher, but because they think they can spend enough to keep it up there and then it will resolve to their profit. At low percentages, someone playing that game has a significant edge, e.g. at 10%, NO bettors would have to out-spend them at 10:1. In a normal market that I believed would resolve to reality, I'm willing to outspend 10:1, or to take as big of a bite as I want and then allow the market to do what it wants - I don't have to keep the price down to profit. But if it's going to resolve-to-market, then I have to bet that my bankroll is 10x theirs or else I lose.
The resolution could be conditional on the market holding steady at a given percentage for a full week.
In other words, it would resolve as soon as the following conditions are met.
It's after Jan 1st 2026
The percentage has not changed for 7 days.
I'll wait to hear everyone else's complaints before I adjust the description...
@Krantz That's also been tried elsewhere; it has a related issue where anyone can always make the percentage change temporarily if they don't want it to resolve at the current value (e.g. because they're in the red). Check out other 'self-resolving' markets around here - a lot of related things have been tried, and as far as I know no one has come up with a satisfactory solution to this problem, though some are more broken than others.
This will only resolve if Eliezer looks at my algorithm.
I knew this might not be quick, or might not happen at all. I know some users don't like long-term markets, especially after loans were removed, but I can't say I didn't know what I was getting into here. I was only surprised by how many YES shares @Krantz was willing to buy well above 2%, but that's not an issue on its own.
I'd be ok with the market staying open, under the original terms. Or an N/A. But changing the market to new criteria does not sit well with me. No matter how well-intended, the suggestions so far are isomorphic to whalebait or other toy markets; if I want to play in a toy market, I'll go find one, I don't need a game retroactively applied to my already-held shares.
FWIW, I actually think the situation with resolution criteria here has parallels to the original question. @Krantz thinks he has a simple solution, proposes it, but it turns out to have some simple flaws. There's no disrespect meant here. I'd like to see prediction markets used more in this way. And God knows we could use more minds working on the alignment problem. I think we'd be friends if we knew each other IRL, you have a lot of interesting knowledge about this topic. But I've watched your videos, read your links, tried to follow the clues you've revealed (bad practice for an 'infohazard'), and I still don't think you've found a solution, and I'm holding more NO shares than anyone else, Ṁ where my mouth is.
I pledge on Kant’s name to try as hard as I can to consent.
This is from the criteria in your other related market. But if you can hold the same standard here, I think you will eventually come around to resolving this NO, with or without further input from EY.
Also, this isn't Rob Miles, correct?
(I'm going to roll a d10 here with the standard terms for plausible deniability) Nope! Just a fan with a similar name. Sometimes I think I might be this Rob though: https://www.explainxkcd.com/wiki/index.php/Rob
Wow. This is incredibly charitable! Thank you for the time you've invested! I would be happy to modify the criteria to resolve N/A on Jan 1st 2026. The only reason I chose partial instead was to avoid being a burden for moderators (I understand they would be required for this and it is recommended to avoid in FAQ). I figured that everyone would expect and be content with the market being bet down to and resolving at 2% thus only distributing my mana to other users.
The goal of this for me, is not to earn mana, but to convert mana into attention on an issue I think is important to think about.
"Krantz thinks he has a simple solution"
Well, sort of. The 'simple' part of the solution is to build an interpretable machine that effectively pays individuals to learn important things (a collective intelligence).
That's a pretty simple solution that I would expect others to be able to see could be a game theoretic way to approach the problem relatively easily (A solution for how to 'teach' the world at scale not to build dangerous systems in the first place). This is what https://manifold.markets/Krantz/krantz-mechanism-demonstration?r=S3JhbnR6 and other predictions were aimed at. If anyone has an argument for why a machine (assuming it worked) wouldn't help that particular situation, I'd love to hear it.
The part I can't expect others to understand, is the architecture for how that could be computationally feasible or scalable. That part is probably going to take several hours of rigorous examination and a decent understanding of the aims of Doug Lenat and Danny Hillis. That part, is what I'd like for @EliezerYudkowsky to look at. I've been trying to just simply get his attention for several years now with no success. I'm typically blocked at every attempt, so I'm kind of out of options. If I can convert a couple hundred dollars worth of mana into a charitable price for some of the people around him to become interested in the approach, maybe their influence will do more than another desperate email from me.
Thanks again.
Eliezer might have a Cheerful Price (https://www.lesswrong.com/posts/MzKKi7niyEqkBPnyu/your-cheerful-price) at which he would happily do the review, even if it would otherwise be an "obvious waste of his time".
Looking over this again after a week away, can I just say that I think everyone is being way too charitable and nice to @Krantz here?
I don't think anyone who's commented except @Krantz would put any substantial probability (above .01, and honestly even .01 seems high) that this would resolve "yes" if @EliezerYudkowsky took a real look. But also there is no reason for him to take a real look because, again, it's just so obviously a waste of his time. But it's also obviously a waste of all of our time too!
The only reason this bet keeps showing up is that @Krantz is single-handedly propping up the "yes" side of the not-very-liquid market, but it's also totally obvious they're not going to take any of the plausible actions that would cause this bet to resolve before 2030. I don't think @Krantz is literally trolling us, but as @ms pointed out, this presents scarily similarly to people who think they've invented perpetual motion machines and are afraid to the share the secret. The epistemics here are terrible and we should stop playing this dumb game. There is nothing to forecast.
@Krantz, you're wasting a lot of people's time propping up this "market."
Your last sentence is incorrect. I decided myself to lose time here, nobody is forcing us to be or trade here. Personslly, i just want to follow how his attempts unfold.
@Krantz why did you pick Eliezer as the only reviewer?
This is disappointing.
Not because it's mean.
Not because it hurts my feelings.
Not because it makes me feel any less confident that this proposition is true.
I actually really wish it did make me feel less confident that this proposition is true.
But that would require new information for me to think about.
You'd have to actually listen to the stuff I've already said, think about it, come up with a good substantive point about it and then share that with me instead. You are not doing that.
You clearly not aiming to do that. You are instead simply repeating your pre-existing assumption that it is not possible for a solution to exist, unless that solution has already been entered into the public domain and been seen/understood by 1,000s of people. That belief has no justification.
"It's totally obvious they're not going to take any of the plausible actions that would cause this bet to resolve before 2030."
If this is true, it is because there exist no plausible actions that would cause this bet to resolve truthfully before 2030. Nobody here to my knowledge here has pointed out any clear actions I can take. Continuing to suggest I should just publish the work openly isn't engaging charitable with what I'm saying.
I thought Manifold was created for individuals to replace all this sort of rhetoric with their money? Isn't the market supposed to determine the legitimacy of predictions? Well, if that's true, then there is a 20% chance right now that I've got a really important paper for Eliezer to read. If that seems wrong, then fix it.
I could understand the concern if you thought I was planning on just resolving the wager and stealing everyone's mana. I am not going to steal your mana. I am going to win your mana. If you are confident enough to risk it.
Thanks. That would be helpful.
The thing I'm trying to accomplish here is to get Manifold users to honestly access their own confidence in whether they should write off my claims before actually examining them.
The way everyone talks, you'd think they were all at least 99% confident.
As far as I can tell, the most confident person is only actually about 80% confident.
@Krantz when you invite people to bet on their belief you did not take into account market duration.
Even if a person is 100% sure this market resolves NO, he would gain only +25% income over 6 years on this market. (The structure of this market implies it can resolve YES early, but cannot resolve NO early)
Such a person could easily find other opportunities with returns after couple weeks.
Another factor is diversification: person can say he is 99% confident, but he still allows the possibility his evaluation is wrong, so he will not bet at all unless there is some not_small margin between price and personal evaluation. (Margin strategy also helps Calibration score a lot).
Third thing: pushing probability to extremes is less profitable than slowly buying at better proces.
20% price DOES NOT mean market evaluates your probability as 20%.
Only with big numbers of markets we can say that IN AVERAGE price equals probability, but we cannot say that about each case.
As far as I can tell, the most confident person is only actually about 80% confident.
Btw, Mirroring your own construction, we get:
"As far as I can tell Krantz is only actually 20% confident".
These are great concerns, thanks.
The market can resolve either 'yes' or 'no' at anytime, if Eliezer wanted it to, but yes I can see why someone might believe he will never look at it and thus the market would just sit here. Funds being tied up is an unfortunate consequence of the system. Do you see any ways around it?
Also, I think extreme bets can have tremendous value when it comes to exploiting market variation. For example, you cite me as saying "I'm 20% confident that I can convince the smartest guy on the planet that he's completely wrong about the thing he's most confident in." That seems like a pretty bold claim.
It seems to me that if there were a rational well calibrated person that thought they had a better than 1% chance of having information that would help us navigate AI, they'd be worth at least listening to, right?
Hopefully you can see why a function like this (a way for researchers to 'bet' on their work before publishing it) would be valuable to society right now. In general, do you think there are any ways I could re-frame this question (Or perhaps convince Eliezer to create a question that allows many researchers to submit/wager on their work)?
What I've been trying to do is help with a path to resolution. Assuming you have a great solution, not publishing prevents us from hyping it which is the only reasonable way Eliezer is going to read it, other than paying him 4-6 figures, imo.
But if you're not going to publish it (narrowly even):
1) do you have a simple self-contained private link (like a Google doc) for the whitepaper? Or a PDF you're willing to email, ready to go now, if he says he'll read it?
2) how long is the doc? This let's Eliezer (and us betting) know the cost of reading it.