LessWrong 2.0 Reader

I feel like one important question here is whether your scenario depends on the assumption that the preferences/demand curves of a consumer are a given to the AI and not changeable to arbitrary preferences.

I think standard economic theories usually don't allow you to do this, but it seems like an important question because if your scenario rests on this, this may be a huge crux.

Let's say we have a bunch of datapoints in $R^n$ that are expected to lie on some lattice, with some noise in the measured positions. We'd like to fit a lattice to these points that hopefully matches the ground truth lattice well. Since just by choosing a very fine lattice we can get an arbitrarily small error without doing anything interesting, there also needs to be some penalty on excessively fine lattices. This is a bit of a strange problem, and an algorithm for it will be presented here.

method

Since this is a lattice problem, the first question to jump to mind should be if we can use the LLL algorithm in some way.

One application of the LLL algorithm is to find integer relations between a given set of real numbers. [wikipedia] A matrix is formed with those real numbers (scaled up by some factor $\zeta$) making up the bottom row, and an identity matrix sitting on top. The algorithm tries to make the basis vectors (the column vectors of the matrix) short, but it can only do so by making integer combinations of the basis vectors. By trying to make the bottom entry of each basis vector small, the algorithm is able to find an integer combination of real numbers that gives 0 (if one exists).

But there's no reason the bottom row has to just be real numbers. We can put in vectors instead, filling up several rows with their entries. The concept should work just the same, and now instead of combining real numbers, we're combining vectors.

For example, say we have 4 datapoints in three dimensional space, $x_i=(x_i,y_i,z_i)$. The we'd use the following matrix as input to the LLL algorithm.

$\left(\begin{array}{cccc}1& & & \\ & 1& & \\ & & 1& \\ & & & 1\\ \zeta x_1& \zeta x_2& \zeta x_3& \zeta x_4\\ \zeta y_1& \zeta y_2& \zeta y_3& \zeta y_4\\ \zeta z_1& \zeta z_2& \zeta z_3& \zeta z_4\end{array}\right)$

Here, $\zeta$ is a tunable parameter. The larger the value of $\zeta$, the more significant any errors in the lower 3 rows will be. So fits with a large $\zeta$ value will be more focused on having a close match to the given datapoints. On the other hand, if the value of $\zeta$ is small, then the significance of the upper 4 rows is relatively more important, which means the fit will try and interpret the datapoints as small integer multiples of the basis lattice vectors.

The above image shows the algorithm at work. Green dot is the origin. Grey dots are the underlying lattice (ground truth). Blue dots are the noisy data points the algorithm takes as input. Yellow dots are the lattice basis vectors returned by the algorithm.

code link

https://github.com/pb1729/latticefit

Run lattice_fit.py to get a quick demo.

API: Import lattice_fit and then call lattice_fit(dataset, zeta) where dataset is a 2d numpy array. First index into the dataset selects the datapoint, and the second selects a coordinate of that datapoint. zeta is just a float, whose effect was described above. The result will be an array of basis vectors, sorted from longest to shortest. These will approach zero length at some point, and it's your responsibility as the caller to cut off the list there. (Or perhaps you already know the size of the lattice basis.)

caveats

Admittedly, due to the large (though still polynomial) time complexity of the LLL algorithm, this method scales poorly with the number of data points. The best suggestion I have so far here is just to run the algorithm on manageable subsets of the data, filter out the near-zero vectors, and then run the algorithm again on all the basis vectors found this way.

...

I originally left this as a stack overflow answer that I came across when initially searching for a solution to this problem.

Linkpost for: https://pbement.com/posts/latticefit.html

Scholars working on safety teams at scaling labs generally selected "working/interning on AI alignment/control"; some of these also selected "working/interning on AI capabilities", as noted. We are independently researching where each alumnus ended up working, as the data is incomplete from this survey (but usually publicly available), and will share separately.

I don't know if I fully get you, but you also nailed it on the head.

In such situation, I think the one weird trick would be to invent a belief system that actively denies being one. To teach people a dogma that would (among other things) insist that there is no dogma, you just see the reality as it is (unlike all the other people, who merely see their dogmas). To invent rituals that consist (among other things) of telling yourself repeatedly that you have no rituals (unlike all the other people). To have leaders that deny being leaders (and yet they are surrounded by followers who obey them, but hey that's just how reality is).

So, basically... science.

Science is the best cult because it convincingly denies being one at all, disguising itself as truth itself. 

I think it's worth admiring science and appreciating it for all the good things it has provided.

And I think it has its limitations, and people should start waking up soon to the fact that if the world is destroyed, humans all destroyed, etc. then science played an instrumental and causal role in that, and part of that is the insanity and evil within that particular cult / worldview. 

https://ezkl.xyz/ 👀

The images appear to be broken.

Or why can't you have a worldview that computes the best answer to any given "what should I do" question, to arbitrary but not infinite precision?

I am not talking about any 'good enough' answer. Whatever you deem 'good enough' to some arbitrary precision.

I am talking about the correct answer every time. This is not a matter of sufficient compute. Because if the answer comes even a fraction of a second AFTER, it is already too late. The answer has to be immediate. To get an answer that is immediate, that means it took zero amount of time, no compute is involved.

Is it something that your reader should be able to infer from this post, or from their own experience of life (assuming they're paying attention?)

Not unless they are Awakened. But my intended audience is that which does not currently walk a spiritual path.

Is this something that you think you know mainly because of your personal experience with pervious worldviews failing you? Some other way?

A mix of my own practice, experience, training, insight, and teachings I've received from someone who knows. 

Merely having worldviews failing you is not sufficient to understand what I am saying. You also have to have found a relative solution to the problem. But if you are sick of worldviews failing you or of failing to live according to truth, then I am claiming there's a solution to that. 

• Sometime in the 2030s, AI will become more of a substitute for human labor, causing an important decline in employment.
• Unemployment will be handled at least as well as the COVID-induced unemployment was handled (sigh). I can hope that AI will enable better governance than that of 2020, but I don't want to bet on when AI will improve governance.

 

It's worse than that, as a whole lot of the unemployment generated by AI is likely to be permanent, and while there are IMO reasonable arguments against AIs taking every job, I don't see a good argument at this time arguing for why this doesn't lead to most human beings being permanently unemployed, other than Roger Dearnley's jobs that focus on people doing surveys to make AI aligned with human values, and I don't see this job class existing for anything more than several months to years.

Hi Clement, I do not have much to add to the previous critiques, I also think that what needs to be simulated is just a consistent enough simulation, so the concept of CI doesn't seem to rule it out.  

You may be interested in a related approach ruling out the sim argument based on computational requirements, as simple simulations should be more likely than complex one, but we are pretty complex. See "The Simplicity Assumption and Some Implications of the Simulation Argument for our Civilization" (https://philarchive.org/rec/PIETSA-6)

Cheers!

I am saying things in a direct and more or less literal manner. Or at least I'm trying to. 

I did use a metaphor. I am not using "poetry"? When I say "Discard Ignorance" I mean that as literally as possible. I think it's somewhat incorrect to call what I'm saying phenomenology. That makes it sound purely subjective.

Am I talking down to you? I did not read it that way. Sorry it comes across that way. I am attempting to be very direct and blunt because I think that's more respectful, and it's how I talk.