mako yass

• mako yass 3 Dec 2022 19:15 UTC

  40 points

  21 ∶ 1

  in reply to: Karthik Tadepalli’s comment on: Re­vis­it­ing EA’s me­dia policy

  The media is an extremely different discursive environment than the EA forum and should have different guidelines.

  I don’t want to assume that the public sphere cannot become earnestly truthseeking, but right now it isn’t at all and bad things happen if you treat it like it is.

• mako yass 22 Feb 2022 7:10 UTC

  38 points

  0 ∶ 0

  on: Me­ta­verse democrati­sa­tion as a po­ten­tial EA cause area

  I think the risk of a VR monopoly is probably low. A metaverse is just an office-compatible VRChat with external app screensharing, and portals. Lots of people are capable of making those. It’s also not apparent to me that Meta are dangerously better at making the hardware than anyone else (apple, HTC and varjo are all stern competitors. Even HP are in the game. Also, by the way, you personally would probably be interested in SimulaVR, they’re about to start making the first portable, self-contained VR workstation computer, and it runs NixOS. It is going to be so so much cooler and better than a laptop. I don’t think I’ll be buying it, it’s only 30 pixels per degree, the maximum perceptible, 60, is right around the corner. But I really hope they stay in business).

  I don’t see any routes to lock people into only using a single service for VR socialization either. You’re concerned about exclusivity with user avatars, having to pay for virtual objects or access to locations, but all of those things make the platform shittier and less useful, if they do that they’ll get outcompeted, there’s nothing they can currently do/​are doing to prevent others from competing with them in that way.

  There’s just way too much competition and incentive for collaboration on standards here. It was heartening to me to read Unreal’s arguments on this. I’d guess that game engines have a big role to play here, and the engines are getting pretty girthy at this point, substituting Unreal out for something made in house is not really going to be feasible for Meta, but Unreal has no interest in aiding the formation of a hardware platform monopoly, so Meta is going to be strongly encouraged to support Unreal’s open standards work.

  I think your political thinking is pretty vague. Democratization and addiction/​misuse/​inhumane tech also have very little to do with each other, if anything they’re aligned. The more democratized the system is, the more addictive systems pop up and compete within it, and the faster their victims find their way to them, one barrier removed.

  Platform monopolies don’t have a greater incentive to promote addiction than an open market already has, and in practice, platforms are usually not the drivers of addiction within their walls. Twitter is a good counter-example, it emerged on the open web, it consisted of some ridiculously crude, barely designed mechanics that worked for reasons nobody understood, and then Twitter went on to become the dominant platform for global discourse, causing a huge amount of damage to the world. The web platform didn’t do that. The openness of the platform didn’t prevent it. Twitter wasn’t addictive because it was carefully designed to be, it just was addictive. And twitter isn’t a monopolistic platform either! It has a simple business model and no lock-in. Your data on twitter is all public. Anyone could take it and make another twitter. You can take it and leave. Sometimes people do leave, but many of them go to mastodon, which is exactly the same, and has the same discourse-political problems.

  The inhumane pattern of twitter persists because addiction is driven by voluntary choices. You can’t solve that sort of problem by adding more liberalization.

  You could only improve the situation by creating platforms that are more engaging, but don’t cause the same harms.

  You need a constructive vision! Give people something better!

  The root cause of the problems are badly designed systems, or an absence of any good designs. People don’t want to design bad systems. They just do whatever pops into their heads.

  I don’t particularly like facebook, but I don’t think they want to destroy democracy or create an unlivable world. No one has those values. I think we need to take care not to get too cynical. Cynicism is a terrible vice. It doesn’t just help you to identify your enemies, it creates enmity and it keeps the rift from ever healing.

  But given the opportunity to lobby Meta, I wouldn’t know what to say to them, yet. Systems determine outcomes but no one is offering a theory of VR hangout and networking systems in relation to how people get information, or have political conversations, or find work, or find love. There are lots of great challenges in designing humane online systems. It’s not obvious that any of the big steps we need to take have much to do with VR. Probably some of them do, but I don’t know what that looks like.

  I am responsible for some visions in social computing (linked: A technology for larger yet more focused communities). Even though they can’t obviously be situated in VR, I’ll mention them just in case they spark any ideas with others.

  Most of my hopes are with venues like Roamresearch, places where it’s as frictionless and accessible as possible to edit linked, structured information collaboratively. Instead of promoting ephemeral, viral content, it promotes longer, continuously evolving content maintained by passionate, curious research and curation communities.

  I think these things could easily evolve into general purpose social network infrastructures too.

  There’s a parallel question there, should we promoting, or, hell, funding, Athensresearch, the open source alternative to Roam? It’s possible that we should! Roam currently doesn’t seem to have plans to support forum and chat primitives, while with Athens it seems like those things can already be added, by anyone, it could end up catering to a much broader, more globally impactful usecase. However, there are some big technical challenges that need to be solved for supporting globally connected versions of these systems, and I don’t really see either project engaging with them yet.

Warn­ings on Weirdness

[Question] Can we ac­tu­ally con­firm that a ve­gan diet is com­pat­i­ble with high in­tel­lec­tual out­put?

• mako yass 14 Jul 2021 1:24 UTC

  29 points

  0 ∶ 0

  on: All Pos­si­ble Views About Hu­man­ity’s Fu­ture Are Wild

  Crazyism about a topic is the view that something crazy must be among the core truths about that topic. Crazyism can be justified when we have good reason to believe that one among several crazy views must be true but where the balance of evidence supports none of the candidates strongly over the others

  Eric Schwitzgebel, Crazyism

• mako yass 22 Apr 2022 23:20 UTC

  25 points

  0 ∶ 0

  on: New: use The Non­lin­ear Library to listen to the top EA Fo­rum posts of all time

  I’m glad this exists.

  Feedback:

  • Why no links to the post in the episode notes? If I find a post interesting, I’m basically always going to want to be able to click through and read comments or vote on it so I need that.

  • I think there shouldn’t be a preamble before reading the title. It kinda destroys the usecase of listening to article titles and skipping to decide whether you want to hear it, it makes that take two or three times as long as it needs to as a result of forcing the listener to sit through the same intro again.
    I’d suggest… just reading the title right away, mentioning the origin, then saying in a different voice, “This audio reading was produced by the Nonlinear Library.” If this has to be here, it’s important for it to be as short as possible. It’d probably be better if it were at the end instead of the beginning. In the vast majority of listens it really doesn’t need to be there at all, given that all of your listeners will either know the origin before they start reading listening, or will listen to enough that they’ll have learned already.

Ap­pendix to Bridg­ing Demonstration

Most so­cial ac­tivity will reside in VR by 2036

• mako yass 22 Jun 2022 2:12 UTC

  15 points

  0 ∶ 0

  on: Im­pact mar­kets may in­cen­tivize pre­dictably net-nega­tive projects

  I should mention that Good Exchange/​impact certs people have discussed this quite a bit. I raised concerns about this issue early on here. Shortly later, I posted the question would (myopic) general public good producers significantly accelerate the development of AGI? to Lesswrong.

  My current thoughts are similar to harsimony’s, it’s probably possible to get the potential negative externalities of a job to factor into the price of the impact cert by having certs take on negative value/​turn into liabilities/​debts if the negative outcomes end up eventuating.
  We don’t know exactly how to implement that well yet, though.

• mako yass 15 Mar 2022 6:51 UTC

  13 points

  0 ∶ 0

  on: EA Pro­jects I’d Like to See

  I think funding good criticism is a really good idea.

  As a meetup organizer, I’m becoming very aware that preserving a culture of criticism is in tension with building a strong social fabric, or making friends. Maintaining the culture is really hard. It would help a bit, to have this very clear signal that we materially value good criticism, and that we protect our critics, even though we’re normally so moderate and agreeable when we meet in person, do not be fooled, we know the value of disagreement too.

  Another thing is, I think this prize would convince a lot of bad critics to work a little harder, and many of them would consequently turn right into good critics, and, I don’t think it harms our culture of criticism to admit that bad critics cause more harm than good. Bad critics misrepresent things, they make everyone who reads them less informed and more confused, they take up time to respond to. It creates noise and rifts that heal slower than they’re torn. I genuinely wouldn’t wish disingenuous critics on anyone, I wouldn’t even wish them on disingenuous critics (alas, by social adjacency).
  And, I think this would cure a lot of them. This proposition that you might be able to deliver a criticism so objectively good that the targets of your ire are committed to paying you for it, officially recognizing that you were right and they were wrong, actioning your advice, and changing. Imagine the level of satisfaction you’d get out of that. The world would be made right.
  A lot of people would be moved by that offer.
  I don’t know if you’d really need to do anything special to keep the criticism good. On some level, people know whether their criticism is going to be useful to the people they’re talking to, they act like it’s deeply ambiguous and it should not be for us (or anyone) to decide whether they’re criticizing in good faith or not, it really isn’t ambiguous. Bad criticism is trivial to identify: You can tell it’s bad because it does not move you and visibly wasn’t intended to. It will seem to be driven by ignorance or intentional misrepresentation. It’s oriented around lowering the target, rather than reforming them. Good criticism shows time and care and you if you’ve ever heard the litany of gendlin then you’ll have no difficulty taking it with relief. Good criticism liberates you from a mistake that you want to stop making. Bad criticism doesn’t.
  Even if there were some rigid, “fair” set of rules forcing you to smile and say thank you and pay for bad criticism, this would not make us any healthier, because bad criticism isn’t actionable, it wasn’t intended to be. It has no use.

  Seeing that only genuinely good criticism can win these prizes, many would be convinced to put down the whip and pick up the scalpel, be more careful in checking their assumptions, citing sources, arguing for the sake of the target rather than some disinterested audience.

• mako yass 14 Dec 2021 6:51 UTC

  13 points

  0 ∶ 0

  in reply to: James Kim’s comment on: Re­sponse to Phil Tor­res’ ‘The Case Against Longter­mism’

  We actually do have a good probability for a large asteroid striking the earth within the next 100 years, btw. It was the product of a major investigation, I believe it was ¹⁄_150,000,000.

  Probabilities don’t have to be a product of a legible, objective or formal process. It can be useful to state our subjective beliefs as probabilities to use them as inputs to a process like that, but also generally it’s just good mental habit to try to maintain a sense of your level of confidence about uncertain events.

[Video] How hav­ing Fast Fourier Trans­forms sooner could have helped with Nu­clear Disar­ma­ment—Veritasium

• mako yass 15 Jan 2023 15:50 UTC

  12 points

  4 ∶ 0

  in reply to: Tegmark’s comment on: [Linkpost] FLI alleged to have offered fund­ing to far right foundation

  I’m curious as to what kind of potentially existentially relevant proposal the NDF would have submitted? What did they think they had to offer?

  (registering a tentative guess: sha256sum ..52ca22c6cd32)

• mako yass 8 Mar 2022 21:13 UTC

  12 points

  0 ∶ 0

  Error

  The value
    NIL
  is not of type
    SIMPLE-STRING
  when binding #:USER-ID163

• mako yass 22 Jun 2022 8:36 UTC

  11 points

  0 ∶ 0

  in reply to: harsimony’s comment on: Im­pact mar­kets may in­cen­tivize pre­dictably net-nega­tive projects

  Crypto’s inability to take debts or enact substantial punishments beyond slashing stakes is a huge limitation and I would like it if we didn’t have to swallow that (ie, if we could just operate in the real world, with non-anonymous impact traders, who can be held accountable for more assets than they’d be willing to lock in a contract.)

  Given enough of that, we would be able to implement this by just having an impact cert that’s implicated in a catastrophe turn into debt/​punishment, and we’d be able to make that disincentive a lot more proportional to the scale of its potential negative externalities, and we would be able to allow the market to figure out how big that risk is for itself, which is pretty much the point of an impact market.

  Though, on reflection, I’m not sure I would want to let the market to decide that. The problem with markets is that they give us a max function, they’re made of auctions, whoever pays most decides the price, and the views of everyone else are not taken into account at all. Markets, in a sense, subject us to the decisions of the people with the most extreme beliefs. Eventually the ones who are extreme and wrong go bankrupt and disappear, but I don’t find this very reassuring, with rare catastrophic risks, which no market participant can have prior experience of. It’s making me think of the unilateralist’s curse.
  So, yeah, maybe we shouldn’t use market processes to price risk of negative externalities.

• mako yass 5 Mar 2022 6:49 UTC

  11 points

  0 ∶ 0

  on: Will “im­pact cer­tifi­cates” value only im­pact?

  I proposed renaming them to “Transferable Attribution”, usually shortened to just “Attribution”. I like this, because the point of selling these things really is to transfer credit, and attribute the act to the buyer.

  This would make it less awkward to refer to fractions of the tokens than “impact certificates”, as “attribution” is a quantity word, rather than a discrete object, you can have amounts of it. Sort of gets rid of the speedbump of having to explain how it’s possible to own a portion of a token or a certificate. (I guess “patronage” does this too, though)

  People seemed to like it. I don’t know what we’d need to do to actually standardize on this name, though.

  It wouldn’t be too catastrophic if we just ended up with a decontextualized “cert”.

• mako yass 1 Mar 2022 1:44 UTC

  11 points

  0 ∶ 0

  on: Be­ing an in­di­vi­d­ual al­ign­ment grantmaker

  Regarding public goods funding, and how it relates to AI Risk: Although I’m really interested in impact certificate markets and I have some mechanisms for it, I’d expect our current designs to make the alignment problem worse, on net.

  The alignment problem is exactly the kind of problem that public goods funding markets we know of will totally mishandle. Progress in AI is largely about knowledge production and improving software infrastructure, which are non-excludable goods (which are generally always public goods), so public goods markets will tend to be really really interested in accelerating progress in AI.
  It’s also in the class of problems where short-term good production lays kindling for long-term negative externalities. Unfortunately, public goods markets are utterly insensitive to that exact class of problem:

  • Investors have short lives, shorter careers, high opportunity cost, a strong preference for an earlier retirement, they are only sensitive to near-term (next 20 years, weighted heavily towards 2 years) returns. If the negative externalities are far away, they wont act on them.

  • Investors are lopsided in how they treat upside and downside. In our current designs, there is no way to expose them to downside risk at all (EG, holding an impact cert cannot make you liable for a debt if the work funded by the impact cert ends up being implicated in bad stuff). Even if we had that, you wouldn’t be able to make the liability large enough to represent the negative externalities that risk-amplifying AI work would produce.

    • With the blockchain primitives that I’m aware of, you can only take debt from quantities of money that the person has staked. On a better blockchain, maybe you could extract the amount of money in their ‘main’ wallet, but that’s also going to be limited. Once you leave blockchain and acquire the incredible legal technologies of a real world state, things become a bit better, you can confiscate their assets, and you can take from their future earnings. But that still gets us nowhere near to tracking the quantity of disvalue that is created by misalignment risk.

      • If we lived in the hell state Cheliax, maybe we could go even further and start enacting disvalue via torture, but even that would likely have a limit, and for most people, who possess compassion, it would introduce further negative externalities, so it would undermine the point, somewhat.

  • It doesn’t matter how many funders take a principled stance and refuse to give payouts for risk-accelerating AI work, if some still will, it’s like all of those principled people just don’t exist, the incentive remains. ~ It’s a max function. It’s biased upwards.

    • If there’s even a chance that someone might fund the work, that attributes non-zero value to investing in risky works’ certs. This probably wouldn’t have much a significant effect on things (I’d expect buyers for certs without some conditional funders’ guarantee of an exit to be quite rare) but it still seems definitely wrong.

    • I guess what I’m trying to get at is, I think funders with different values/​worldviews really need some sort of coordination mechanism for negotiating about externalities. Maybe a mechanism like that could solve some of the existing problems with positive externalities too (currently subject to free-rider issues). Public goods markets should be all about tracking externalities, so I don’t think this is too much to hope for.

    • I would guess that most large software public goods funders active right now are not longtermists, and I sort of don’t expect them to ever universally recognize the alignment problem, but it’s conceivable that could change.

  The best solutions I can offer right now:

  • Talk about the issue

    • Make sure everyone who wants to make an impact cert market knows about this long-term negative externality tracking problem and is willing to integrate more sophisticated systems that don’t have this problem, as we think of them.

    • Raise awareness of the existence of the various short-term goods that are long-term bads, so that funders who use these systems hear about them.

  • I don’t like this, but we might not have anything better for now: Introduce some sort of human-judgement filter that prevents impact certs from being minted if they have untrackable negative externalities, that is, if they have negative externalities that exceed what the system can counterbalance through fines or debt-taking. Just don’t allow that kind of work into the market.
    There might be much more elegant, decentralized ways of doing that. I don’t have any today.

  What links here?

  • Would (my­opic) gen­eral pub­lic good pro­duc­ers sig­nifi­cantly ac­cel­er­ate the de­vel­op­ment of AGI? by mako yass (LessWrong; 2 Mar 2022 23:47 UTC; 25 points)
  • mako yass's comment on Im­pact mar­kets may in­cen­tivize pre­dictably net-nega­tive projects by Ofer (22 Jun 2022 2:12 UTC; 15 points)
  • Jackson Wagner's comment on Will “im­pact cer­tifi­cates” value only im­pact? by RyanCarey (5 Mar 2022 3:30 UTC; 13 points)
  • Ofer's comment on Be­ing an in­di­vi­d­ual al­ign­ment grantmaker by A_donor (6 Mar 2022 4:34 UTC; 1 point)

NLNet, open source soft­ware fun­ders, are hiring as­ses­sors. I think their work could be im­por­tant.

• mako yass 5 Dec 2022 21:08 UTC

  10 points

  2 ∶ 0

  in reply to: John_Maxwell’s comment on: Re­vis­it­ing EA’s me­dia policy

  It is a joke, but it’s an appropriate one.

  EA has a pathology of insisting that we defer to data even in situations where sufficient quantities of data can’t be practically collected before a decision is necessary.

  And that is extremely relevant to EA’s media problem.

  Say it takes 100 datapoints over 10 years to make an informed decision. During that time:

  • The media ecosystem, the character of the discourse, the institutions (there are now prediction markets involved btw) and the dominant moral worldviews of the audience has completely changed, you no longer need the answer to the question this data answers.

  • You have already been assassinated for not engaging in a principled and decisive way.

  I should assume that I’m talking to someone who has this pathology and needs me to explain what the alternative to “defer to data” even is: Get better at interpreting the data you already have. Seek theories of communication that’re general enough and robust enough that you don’t strictly need to collect further data to validate them. Test them anyway, but you can’t wait for the tests to conclude before deploying.

• mako yass 28 Mar 2022 2:17 UTC

  9 points

  0 ∶ 0

  in reply to: Dawn Drescher’s comment on: Will “im­pact cer­tifi­cates” value only im­pact?

  Can you imagine a way to get a person to engage well with an impact market (or any market) when they don’t understand money/​beneficial self-reifying games or whatever?