LessWrong 2.0 Reader

I'd say "in many contexts" in practice refers to when you are already working with relatively blessed information. It's just that while most domains are overwhelmingly filled with garbage information (e.g. if you put up a camera at a random position on the earth, what it records will be ~useless), the fact that they are so filled with garbage means that we don't naturally think of them as being "real domains".

Basically, I don't mean that blessed information is some obscure thing that you wouldn't expect to encounter, I mean that people try to work with as much blessed information as possible. Logs were sort of a special case of being unusually-garbage.

You can't distill information in advance of a bug (or anomaly, or attack) because a bug by definition is going to be breaking all of the past behavior & invariants governing normal behavior that any distillation was based on.

Depends. If the system is very buggy, there's gonna be lots of bugs to distill from. Which bring us to the second part...

The logs are for the exceptions - which are precisely what any non-end-to-end lossy compression (factor analysis or otherwise) will correctly throw out information about to compress as residuals to ignore in favor of the 'signal'.

Even if lossy compression threw out the exceptions we were interested in as being noise, that would actually still be useful as a form of outlier detection. One could just zoom in on the biggest residuals and check what was going on there.

Issue is, the logs end up containing ~all the exceptions, including exceptional user behavior and exceptional user setups and exceptionally error-throwing non-buggy code, but the logs are only useful for bugs/attacks/etc. because the former behaviors are fine and should be supported.

I've been tracking the Rootclaim debate from the sidelines and finding it quite an interesting example of high-profile rationality.

Would you prefer the term "high-performance rationality" over "high-profile rationality"?

Hello everyone! My name is Roman Maksimovich, I am an immigrant from Russia, currently finishing high school in Serbia. My primary specialization is mathematics, and back in middle school I have had enough education in abstract mathematics (from calculus to category theory and topology) to call myself a mathematician.

My other strong interests include computer science and programming (specifically functional programming, theoretical CS, AI, and systems programming s.a. Linux) as well as languages (specifically Asian languages like Japanese).

I ended up here after reading HP:MOR, which I consider to be an all-time masterpiece. The Sequences are very good too, although not that gripping. Rationality is a very important principle in my life, and so far I found the forum to be very well-organized and the posts to be very informative and well-written, so I will definitely stick around and try to engage in the forum to the best of my ability.

I thought I might do a bit of self-advertising as well. Here's my GitHub: https://github.com/thornoar

If any of you use this very niche mathematical graphics tool called Asymptote, you might be interested to know that I have been developing a cool 6000-line Asymptote library called 'smoothmanifold', which is sort of like a JavaScript framework (an analogy that I do not like) but for drawing abstract mathematical diagrams with Asymptote, whose main problem is the lack of abstraction. In plain Asymptote, you usually have to specify all the coordinates manually and draw objects line by line. In my library I make it so that the code resembles the logical structure of the picture more. You can draw a set as a blob on the plane, and then draw arrows that connect different sets, which would be a nightmare to do manually. And this is only the beginning -- there is a lot more features. If any of this is what interests you, feel free to read the README.md.

I have also written some mathematical papers, the most recent one paired with a software program for strong password creation. If you are interested in cryptography and cybersecurity, and would to create strong passwords using a hashing algorithm, you can take a look at 'password-hash', which contains both the algorithm source and binaries, as well as the paper/documentation.

Consider the teleporter as a machine that does two things: deconstructs an input i and constructs an output o. 
If you divide the machine logically into these two functions, d and c, which are responsible for deconstructing and constructing respectively, you have four ways the machine could function or not function:

If neither d or c work, the machine doesn't do anything. 

If d works but c doesn't, the machine definitely kills or destroys the input person. 

If d doesn't work and c does, the machine makes a copy of the person. If a being walked into the machine and found that this happened, the input being would be in my opinion justified in saying that they oppose being deconstructed.

If d works and c works, then we have a functioning teleporter. This is similar to the previous situation, just with "being i" destroyed. I find it hard to believe this is preferable in some way from the perspective of the input being. 

I think there is possibly a good argument we should accept that this leads to some sort of nihilism about the value / coherence of our existence as discrete individuals, but personally, I maintain too much uncertainty to be okay with stepping into a "teleporter" type system that is more novel than going to sleep (which does after all destroy the being that goes to sleep and create a being that wakes up)

How scarce are tickets/"seats"?

COMMENT THREAD

If you comment anywhere other than here, Nathan will delete your comment.

This all seems very teleological. Do you have thoughts on what the teleology of the universe could be under this model? 

One optimal predictor for this data would be to maintain a belief over which of the three HMM states we are in

As well as inferring the HMM itself from the data.

Features to benefit people accused of X may benefit mostly people who have been unjustly accused.  So looking at the value to the entire category "people accused of X" may be wrong.  You should look at the value to the subset that it was meant to protect.

Personally, I like mentally splitting the space into AI safety (emphasis on measurement and control), AI alignment (getting it to align to the operators purposes and actually do what the operators desire), and AI value-alignment (getting the AI to understand and care about what people need and want). Feels like a Venn diagram with a lot of overlap, and yet some distinct non-overlap spaces.

By my framing, Redwood research and METR are more centrally AI safety. ARC/Paul's research agenda more of a mix of AI safety and AI alignment. MIRI's work to fundamentally understand and shape Agents is a mix of AI alignment and AI value-alignment. Obviously success there would have the downstream effect of robustly improving AI safety (reducing the need for careful evals and control), but is a more difficult approach in general with less immediate applicability. I think we need all these things!