Archimedes

Karma: 382

Archimedes 1 Dec 2020 0:44 UTC
3 points
on: Syntax, semantics, and symbol grounding, simplified
I’d like to nitpick a bit with the E, A, R example. In particular, a Rosetta Stone R is far from an equivalence between the features of A and B. It’s a tiny subset of features of the world expressed in both Linear A and in English. The trick is that this tiny hook between languages can be bootstrapped into an imperfect relationship between the symbology of Linear A and the symbology of English up to the limits of indeterminacy. In a sufficiently large corpus of Linear A, a bootstrapping R isn’t even needed. Nearly all of the information for translation is contained within E and the fact that E describes a shared world. In some sense, it’s self-grounded in a similarly messy and largely self-referential way that any agent’s command of any language is grounded.
With regard to GPT-n, I don’t think the hurdle is groundedness. Given a sufficiently vast corpus of language, GPT-n will achieve a level of groundedness where it understands language at a human level but lacks the ability to make intelligent extrapolations from that understanding (e.g. invent general relativity), which is rather a different problem.
To properly address “GPT-n series of algorithms will not reach super-human levels of capability”, I need to understand more clearly what you mean by “capability”. Capability of what in particular? Can you describe a test that could verify superhuman capability? Is this different than a Turing test?

Archimedes 19 Feb 2021 15:18 UTC
1 point
in reply to: MikkW’s comment on: Your Cheerful Price
I really like the idea of this sort of currency but I feel like if it’s easily fungible for dollars, then it loses power as a separate concept and is subject to the market forces of Wall Street trying to make $$$ from manipulating it some way or another.
Of course, whenever multiple currencies exist, there will always be exchanges whether official or black market. However, I think there should be careful thought about what sorts of friction or limitations should be in place to preserve the intent of the currency.
Maybe you only allow it to be associated with real individual people and possibly have caps on balances or transactions but maybe that would make things worse somehow. Hard to say without serious study.

Archimedes 19 Feb 2021 16:04 UTC
2 points
on: Your Cheerful Price
I think this is a great concept generally. However, I can’t think of many situations where I’d personally be able to come up with a “cheerful price” where I’d be happy to do something that I wouldn’t do for free.
For example, if a friend asks me to bake a cake or drive them to the airport, I wouldn’t be more cheerful doing that for, say, $1000 than for free. I’d probably be less cheerful since I know that’s a ridiculous price, even if they wanted to pay it. Maybe that’s just because I have enough money that a one-off extra $1000 wouldn’t have a meaningful impact on how I live my life or maybe it’s just that there isn’t a strong connection between dollars and my mental state.
I can imagine situations where I would have a “cheerful price” but these mostly involve doing something for a stranger that doesn’t take much time or effort. For anything that requires significant time or effort, I’d have to activate my executive functioning to try to balance things rationally. Ultimately, I think money is too abstract to be a good motivator for me. My monkey brain needs some help to translate it into something more tangible than something that merely has the potential to be exchanged for goods and services at some later point.

Archimedes 5 Mar 2021 1:05 UTC
4 points
in reply to: Caridorc Tergilti’s comment on: Fun with +12 OOMs of Compute
With AB pruning, the branching factor is closer to 2 for Stockfish.

Archimedes 5 Mar 2021 20:38 UTC
1 point
in reply to: Jameson Quinn’s comment on: A whirlwind tour of Ethereum finance
>And as far as I can tell, non-proof-of-work cryptocurrency is mostly a mirage or even a deliberate red herring / bait-and-switch.
How familiar are you with Ethereum’s plan to transition to Proof of Stake? It’s definitely not there yet but making (slow) progress.

Archimedes 22 Mar 2021 20:43 UTC
3 points
in reply to: Eric Raymond’s comment on: Open, Free, Safe: Choose Two
I think there is an important distinction between being “safe” from ideas and “safe” from interpersonal attacks. In an online space, I expect moderation to control more for the latter than the former, protecting not against wrongspeech so much as various forms of harassment.
Rational discourse is rarely possible in an environment that is not protected by moderation or at least shared norms of appropriate communication (this protection tends to be correlated with “openness”). Having free speech on a particular platform is rarely useful if it’s drowned out by toxicity. I support strong freedom of ideas (within the bounds of topicality) but when the expression of those ideas is in bad faith, there is no great value in protecting that particular form of expression.
There is a hypothesis that that unconstrained speech and debate will lead to wrong concepts fading away and the less wrong concepts rising to more common acceptance but internet history suggests that this ideal can’t last for long in a truly free space unless that freedom is never actually tested. As soon as bad actors are involved, you either have to restrict freedom or else experience a degradation in discourse (or both). If safety is not considered, then a platform effectively operates at the level of its worst users.

Archimedes 23 Mar 2021 13:57 UTC
6 points
on: Demand offsetting
There’s a bit of a wrinkle here in that humane eggs aren’t completely fungible with non-humane eggs. Healthier and happier chickens tend to produce better quality eggs that are distinguishable from factory-farmed eggs. This approach seems plausible for offsetting demand but leaves the consumer with an inferior product at a higher cost. This isn’t a big deal for a croissant but would be for, say, an omelet.

Archimedes 23 Mar 2021 14:56 UTC
1 point
on: Trapped Priors As A Basic Problem Of Rationality
The key to fixing a phobia or changing closely held beliefs seems to involve addressing it within an environment that is perceived as safe. When a person feels attacked or threatened, the instinctive and emotional System 1 of the brain tends to dominate and this system is more heuristic than Bayesian. You can’t rationally update a prior that is not based on rationality, so the trick is to coax that prior into a space where it can be examined by the more logical and deliberate System 2.
The cognitive vs emotional section addresses this but I’m not convinced that the set of purely epistemic trapped priors is significant compared to the much larger set of emotionally trapped priors. Are there better examples than the polar bear of non-emotional trapping?

Archimedes 23 Mar 2021 21:23 UTC
2 points
in reply to: Pongo’s comment on: Demand offsetting
Sure, but the non-fungibility reintroduces the distribution and packaging complications. Part of the premise was to “sell these on the wholesale market as if they were totally normal eggs.” The decoupling only works under fungibility assumptions, i.e., when the quality difference is not distinguishable or not important. Fortunately, this is close enough to true for many parts of the market and a partial solution is still progress.

Archimedes 25 Mar 2021 22:31 UTC
6 points
on: The best frequently don’t rise to the top
The idea that better quality cooking corresponds to higher viewership is only a rough expectation even when all else is held equal, which never actually happens. YouTube popularity is highly memetic and subject to algorithms, network effects, and plain dumb luck. The factors involved in success are numerous and difficult to predict, but some of the most important factors are quality content, posting regularity, high entertainment value, viewer engagement, and having an angle or niche to stand out. Being the best cook in the world guarantees not a single one of these.

Archimedes 29 Mar 2021 16:57 UTC
1 point
in reply to: Eric Raymond’s comment on: Open, Free, Safe: Choose Two
As long as people are dishonestly gaming the system, there will always be problems and there is no silver bullet solution. It’s a fundamentally hard problem of balancing competing values. Any model proposed will have failings. The best we can do is to try to balance competing values appropriately for each individual platform. Each one will have different tilts but I doubt rejecting safety entirely is likely to be a good idea in most cases.
It’s often tempting to idealize one particular value or another but when any particular value is taken to an extreme, the others suffer greatly. If you can back away from a pure ideal in any dimension, then the overall result tends to be more functional and robust, though never perfect.

Archimedes 29 Mar 2021 21:42 UTC
9 points
in reply to: NaiveTortoise’s comment on: Core Pathways of Aging
From https://www.livescience.com/61568-naked-mole-rats-no-aging.html:
In the lab, the cause of death is usually hard to find; the main issue that shows up in necropsies, Buffenstein said, are mouth sores, indicating the animals weren’t eating, drinking or producing saliva well in their last few days and infection set in.
“We really don’t know what’s killing them at this point,” Buffenstein said.

Archimedes 29 Mar 2021 22:26 UTC
5 points
in reply to: AnnaSalamon’s comment on: Toward A Bayesian Theory Of Willpower
To handle this, it would probably work to extend the “motionless” prior to a more abstract “status quo” prior. It’s not necessarily related to physical movement (it could even be the opposite if it were a sports group instead of a study group in your example) but rather an aversion to expending effort or experiencing potential undesired consequences. The prior says, “Things are fine. It’s not worth going out of the way to change that.”
With this extended version, creativity-requiring tasks are analogous to physically-demanding tasks. With pre-scripted tasks, it’s not as demanding in the moment.

Archimedes 29 Mar 2021 22:50 UTC
3 points
in reply to: Kaj_Sotala’s comment on: Toward A Bayesian Theory Of Willpower
Under the bidding system theory, if the non-winning bids still have to pay out some fraction of the amount bid even when they lose, then bidding wars are clearly costly. Even when the executive control agent is winning all the bids, resources are being drained every auction in some proportion to how strongly other agents are still bidding. This seems to align with my own perceptions at first glance and explains how control wanes over time.

Archimedes 1 Apr 2021 16:00 UTC
5 points
in reply to: Liam Goddard’s comment on: Why We Launched LessWrong.SubStack
Maybe posts like “The Solution to the Hard Problem of Consciousness” don’t actually exist other than as a joke title?

Archimedes 12 Apr 2021 16:47 UTC
1 point
on: Bayes’ theorem, plausible deniability, and smiley faces
With regard to the smiley face problem, you can keep the other signifiers of tone and just phase out the emoticon so that it’s “That’s awesome. Thanks!” rather than a more passive “thanks”.

Archimedes 6 May 2021 16:00 UTC
4 points
on: Could MMRPGs be used to test economic theories?
There are some pretty major differences between game economies and real life that would need to be carefully considered.
1. Motivation: People generally play games voluntarily for purposes of enjoyment and can stop when they get bored or fed up. Giving up on real life is entirely different. Games closer to reality will likely suffer from loss of interest.
2. Difficulty: Surviving and thriving in most MMRPGs is much simpler than reality where (for most people) huge amounts of time and energy are spent earning money for the basics of living (rent, food, bills) and engaged in bodily necessities like eating and sleeping. Games without serious consequences for failing to meet basic monetary and biological needs will miss huge categories of economic incentives (e.g. wage slavery).
3. Growth: Games tend to have monotonically increasing levels/stats for player characteristics. Reality is so much more frustrating. You lose strength if you stop training or get injured. You forget things you’ve learned, especially if you don’t practice. Growth in the real world is longer and harder and naturally decays over time even when there aren’t major regressions like an injury. Real-world growth tends to be asymptotically limited by individual aptitudes determined by things like the genetic lottery and easily hampered by arbitrary local environment and corresponding access to resources.

Archimedes 6 May 2021 21:08 UTC
7 points
in reply to: StyleOfDog’s comment on: Your Dog is Even Smarter Than You Think
My priors include the idea that both animal intelligence is not that different from humans and also that humans tend to overly anthropomorphize animal cognition. The biggest misunderstandings of animal cognition are much like misunderstandings humans have of foreign cultures, often involving forms of mind projection fallacies where we assume other’s values, motivations, priorities, and perceptions are more similar (or more different) than is justified.

Archimedes 2 Jun 2021 20:58 UTC
1 point
AF
on: My Current Take on Counterfactuals
I’m having trouble accepting that the Troll Bridge scenario is well-posed as opposed to a Russell-like paradox. Perhaps someone can clarify what I’m missing.

In my mind, there are two options:
1. If PA is inconsistent, then math is in ruins and any PA-based reasoning for crossing the bridge could be inconsistent and the troll blows up the bridge. Do not cross.
2. If PA is consistent, then the agent cannot prove U = −10 (or anything else inconsistent) under the assumption that the agent already crossed, and therefore Löb’s theorem fails to apply. In this case, there is no weird certainty that crossing is doomed.
Now until/unless PA is proven inconsistent, it’s reasonable to assign the majority of probability mass to the prior that PA is, in fact, consistent and we can ignore counterfactuals that depend on proving otherwise since if that’s proven, none of the rest of the reasoning matters anyway until foundational logic has been reformulated on a consistent basis.

Archimedes 14 Jun 2021 15:48 UTC
2 points
in reply to: abramdemski’s comment on: My Current Take on Counterfactuals
I’m objecting to an assumption that contradicts a previous assumption which leads to inconsistent PA. If PA is consistent, then we can’t just suppose 1 = 2 because we feel like it. If 1 = 2, then PA is inconsistent and math is broken.

Assuming PA is consistent and assuming the agent has crossed, then U for crossing cannot be −10 or the agent would not have crossed. Weird assumptions like “1 = 2” or “the agent proves crossing implies U = −10″ contradict the existing assumption of PA consistency. Any proof that involves inconsistency of PA as a key step is immediately suspect.

Let’s take a look at the proof using a simpler assumption.

Suppose the agent crosses. Further suppose the agent proves crossing implies 1 = 2. Such a proof means PA is inconsistent, so 1 = 2 is indeed “true” within PA. Thus “agent proves crossing implies 1 = 2″ implies 1 = 2. Therefore by Löb’s theorem, crossing implies 1 = 2.

This is just a fancy way of abusing Löb’s theorem to bootstrap a vacuously true statement by applying it to an inconsistent axiom system. Within consistent PA, this proof is nonsense. With inconsistent PA, every proof in PA is nonsense, so it doesn’t make sense to make assumptions that lead to inconsistent PA. If the proof leads to “PA is inconsistent”, then every step that follows from that step is unreliable.

In my opinion, the whole Troll Bridge Problem boils down to supposing False = True and then trying to apply logical reasoning to the resulting inconsistent system. Of course you get a paradox when you suppose U = −10 and U >= 0 simultaneously.