LessWrong 2.0 Reader

More on green errors, I think they do exists. There is a difference between an invasive species and a predator. Green probably allows for predators easier than white or black that would call them murderes. But being disruptive to the harmony is an actual violation green registers.

Imagine you have a snake problem in your houses yard. You could get angry and kill every snake you see (haphazard, random and laboursome the red way to address it). You could poison your yard (but then your flowers might die or your food supply gets fouled, the black way). For wheel completeness sake, wall (white way) or scarecrow (blue way). Or you could introduce a predator species that eats snakes (the green way). Even if the effect is to diminish a component you address it by constructing more components (add species). And likely when the problem is "solved" the predator and prey are in balance and in a way the snakes existence functions as a foundation for the foodchain for the predator.

The hard thing about green as it is the anti-color of the agent color black it doesn't engage in problem solving. Nature by itself is a defenceless victim. People who care about nature and are naturalistic are a bit different thing. In making a choice what "harmony" you are defending you are probably injecting somewhat of a agentic subjective choice.

This reads to me about magic the gathering terms how it's useful to be green over being black, althought it mostly uses black terminology.

The violin parenting situation is a good example how black is about certain costs for uncertain benefits and green is about foregoing certain advantages to avoid uncertain disadvantages.

Black is very local, me, this game, this battle and green is very global, ecosystem, species, bordering beyond the unknown.

The linked article had more focus on laziness where it was more anti-black where blue and green attitudes are somewhat mixed up. It is very uncharastic of blue ot have "??????" steps in the plan and in general less blue shades. I do think the word "medicority" is a very black/blue way of characterising the proprerty as would be expected when the word is derived to be an antonym for their core concepts. I think actually the part where you get to the middle is more active, green wants you to "grow up" and "go be you" but it doesn't like going over carrying capacity or attempt things not in your nature. In away blue doesn't bother to the same extent as illusion and visions are adequate and when black cuts something out it cuts it out to 0 if not negative.

I could for example think that green calls "corruption" what black would call "excellence" or "expertise" and "minimal" what blue calls "optimal". It would be interesting what would be the green concept for the green ideal but my intuition seems to be weaker, maybe "vibrant" (seems to capture diversity well but is not so directly a counterpart to mediocrity)?

More generally, the whole setup is just 1) train your model in a sandbox environment 2) have the weights of your model fixed during deployment.

I think this is missing a crucial point of Counterfactual Oracles (as implemented in ML). I came upon Paul Christiano's Counterfactual oversight vs. training data today which explains this very well. (What he calls Counterfactual Oversight is basically Counterfactual Oracles as applied to predicting/imitating humans.) The problem with the standard supervised learning system that you're describing is:

The problem is not stationary, and over time the training data becomes less relevant. There are spurious correlations in the training data that don’t generalize to the test data.

And Counterfactual Oracles (in ML terms) is actually an attempt to solve these problems:

Counterfactual oversight consists of labelling a random subset of data and using it as online training data. The key difference is that any given data point may become a training data point, with the decision made after the learning system has made a decision about it. As long as the randomization is unpredictable to the learner, this gives us a formal guarantee that there can’t be any noticeable difference between the training and test data. And therefore if our learner behaves well on training data, it really must behave well on test data.

Variable truth-value (VTV) of a sentence is a technical thing in formal semantics - it means that the truth-value of this sentence depends on the little thingy called variable assignment. While the term might seem misleading, it is useful for explaining why we still claim "He walked in" has a truth-value - it first has the VTV, and then we find some "discourse" assignment that converts VTV to truth-value. Also, variable assignment can be manipulated from within the sentence (anaphora, movement, you name it).

With an ad blocker, I see almost nothing on the page. No, I am not going to turn it off.

"Believe it or not, for some decades, there was a serious debate about whether people really had mental images in their mind—an actual picture of a chair somewhere—or if people just naively thought they had mental images (having been misled by "introspection", a very bad forbidden activity), while actually just having a little "chair" label, like a LISP token, active in their brain" - AFAIK, you misrepresent the debate. It was rather about what is primary and what is secondary. Sure, your brain paints a chair - but does it first paint the chair and then search for its properties, or is the mental image merely one of the properties of a pre-found concept (the correspondence to the options you represent is _in the order given_)? Not _that_ silly, is it? (It still has a right answer, but it happens to be the second one not the first one.)

I posted on Facebook, and LW might actually also be a good place for some subset of topics.

What is wrong (if you don't try to match the real biological taxonomy of Pisces) with definition like "vertebrate constantly living in water" for "fish"? While true_fish and dolphins have some differences, they have the very common points that led to the unification in the first place.

This doesn't exactly apply because most welfare spending (and in fact most government spend) is leveraged against future economic growth and inflation, so it actually is possible to create self-fulfilling prophecies via stimulus spending so long as that spend eventually produces real growth. This case might or might not be the most efficient way to do that, but governments are rarely trying to maximize efficiency at the expense of all else so we shouldn't really count that against the argument. So this would seem a neutral, at worst, scenario rather than one of dead-weight loss as you argue.

In one scenario, the political leaders of the two towns decide to unite the two towns. Nothing except the name will change. So, there is now a town C that has 1010 inhabitants. Will town C have a functional economy? Of course! After all, all that was done was subsuming two already functional economies.

I feel a bit wary of using this as an axiom for further thinking about immigration between countries. I've not really thought about this issue before, but my guess is if two countries with very different levels of economic development open their borders to each other this could be fairly damaging. It might substantially increase prices and demand of land in the less economically developed country, causing upheaval (e.g. poorer people in the more developed country buying out middle class people in the less developed country); it may damage a lot of their industries where the more economically developed country has more efficient abilities. I also think, especially if there's big cultural and language barriers that a lot of basic norms for interacting with strangers and such can be very confused and cause damage in the years/decades it takes to integrate.

I don't have much of an opinion here, just thought I'd mention that I got off the bus a bit at this part of the argument.