Posts
Comments
I think this is a good object level post. Problem is, I don't think MIRI is at the object level. Quote from the comm. strat.: "The main audience we want to reach is policymakers."
Communication is no longer a passive background channel for observing a world, but speech becomes an action changing it. Predictions start to influence the things they predict.
Say AI doom is a certainty. People will be afraid, and stop research. Few years later doom doesn't happen, everyone complains.
Say AI doom is an impossibility. Research continues, something something paperclips. Few years later nobody will complain because no one will be alive.
(This example itself is overly simplistic, real-world politics and speech actions are even more counterintuitive.)
So MIRI became a political organization. Their stated goal is "STOP AI", and they took the radical approach to it. Politics is different from rationality, and radical politics is different from standard politics.
For example, they say they want to shatter the overton window. Infighting usually breaks groups; but during that, the opponents need to engage with their position, which is a stated subgoal.
It's ironic that a certain someone said Politics is the Mind-Killer a decade ago. But because of that, I think they know what they are doing. And it might work in the end.
How to be the smartest person in the world:
Let's say A is smarter than B if A knows about topic X, but B doesn't.
Step 1: Let's say you don't know about quantum biology. But Charlie knows, because they currently do a phd in it.
Step 2: Go to Charlie. Say: "I heard about quantum biology, and it sounds interesting. Could you give quick intro on it?"
Step 3: Charlie says (eager to talk about the cool idea they found): "Sure. Quantum biology is [two hour forty-seven minute long monologe]."
Step 4: Important! Listen to it.
Step 5: BOOM! Now you also know about quantum biology.
Repeat it for every topic. If you partition humanity along every topics, you will always be in the in-the-know part. By Zorn's lemma[1] you will be one of the smartest person in the world.
To be fair, in real life there are time and energy bounds, not everyone has time to talk about their topic, and active listening can be a hard mental work. But it worked for me a surprising amount of times. Well, surprising at first, then I adjusted my expectations.
- ^
actually you might not need Zorn's lemma for this, but it sounds so cool
Taking a bad option away might be worse for a person, but will be much better for the people. These regulations (no selling organs or sex) exists, becuse in a free market there would be a race-to-bottom which would not increase human values.
Suppose we allow selling sex for rent. The number of rentable apartmants stays the same; however, there will more demand for them, because some people can now pay for them by non-monetary means. Because of this, the rent prices will increase, and that would just accelerate the rent-problem.
While exchanging kidneys for medical treatment is OK for me, it should not be mixed with the standard money market. The forces of money markets usually optimize for dollar value, which could be decoupled from human wellbeing. The result would be a worse state for everyone.
Also: If the rent is so high, why can't a developer build a new complex? They could rent it out and would very fastly pay for itself. It would increase the number of flats and lower rents. These bad options try to solve a supply issue from demand side.
Searched PsyNet on Google, and I think PSYNet refers to the netcode for RocketLeague, a popular game. Maybe they pulled text message logs from somewhere; based on the "ForgeModLoader" token, it's plausible.
Alternative guess is this, a python library for online behavioural experiments. It connects to Dallinger and Mechanical Turk.
On Google, the string "PsyNetMessage" also appeared in this paper and at a few gpt2 vocab lists, but no other results for me.
On Bing/DuckDuckGo it outputted a lot more Reddit threads with RocketLeague crash logs. The crash logs are full of messages like [0187.84] PsyNet: PsyNetRequestQue_X_1 SendRequest ID=PsyNetMessage_X_57 Message=PsyNetMessage_X_57
, so I guess it's an RL (as in RocketLeague) thing. It was also found in some (clearly) GPT-generated texts.
The world is a complicated and chaotic place. Anything could interact with everything, and some of these are good. This post describes that general paralysis of the insane can be cured with malaria. At least if they do not die during the treatment.
If late-stage syphilis (general paralysis) isn't treated, then they probably die 3-5 years with progressively worse symptoms each year. So even when 5-20% of the died immediately when the treatment started, they still had better survival rates in one and five years. A morbid example of an expected value choice: waiting for a certain long death vs taking a chance at a short or longer possible lifetime.
If they were allowed to choose at all, where the "they" means the patients. The post mentions that Wagner-Jauregg maybe hasn't asked for consent when he tried his experiments. But this is on par for the age, early XX. century hasn't considered mentally ill patients human. Anyway, at this point, I disagree with the tone of the post, which may support human experimentation without consent. I mean, the guy just tried a bunch of diseases on terminally ill because of a fight-fire-with-fire theory and randomly found one which somewhat works.
He got a Nobel for this discovery, and a few years later he supported eugenics and anti-Semitism. Nowadays we don't use it because somebody else discovered penicillin and half of medicine was solved. We know a bit more about malaria. We don't know why this therapy worked and other high-temperature methods don't. The guy got a few places named after him in Austria.
The article is well-researched. Does it carve reality at its joints? I don't feel like it describes a reliable and ethical scientific process. But maybe sometimes you just can't, because the world is a complicated and chaotic place.
What does this post add to the conversation?
Two pictures of elephant seals.
How did this post affect you, your thinking, and your actions?
I am, if not deeply, but certainly affected by this post. I felt some kind of joy looking at these animals. It calmed my anger and made my thoughts somewhat happier. I started to believe the world can become a better place, and I would like to make it happen. This post made me a better person.
Does it make accurate claims? Does it carve reality at the joints? How do you know?
The title says elephant seals 2 and contains 2 pictures of elephant seals, which is accurate. However, I do not think it carves reality because these animals don't have joints. I know it from experimental evidence: I once interacted with a toy model of a seal and it was soft and fluffy and without bones.
Is there a subclaim of this post that you can test?
no
What followup work would you like to see building on this post?
You wouldn't guess it, but I have an idea...
I was always surprised that small changes in public perception, a slight change in consumption or political opinion can have large effects. This post introduced the concept of the social behaviour curves for me, and it feels like explains quite a lot of things. The writer presents some example behaviours and movements (like why revolutions start slowly or why societal changes are sticky), and then it provides clear explanations for them using this model. Which explains how to use social behaviour curves and verifies some of the model's predictions at the same time.
The second half of the post bases a theory on what an ideal society would look like, and how should you act on a radical-conformists axis. Be a radical except if only radicals are around you is a cool slogan for a punk band, but even he writes that he's gonna do a little trolling. I feel like there are some missing assumptions about why he chooses these curves.
In the addendum, there are references to other uses in the literature, which can be used as a jumping point for further understanding. What I'm missing from this post is the discussion of large networks. Everyone knows everyone in a small group, but changes propagate over time for large ones; it also matters if someone has few or many connections. There is also some kind of criticality in large networks too, but it's a bit different. Also, the math gets much more complicated, in fact, graph criticality results are few and hard, and most places use computer simulations instead of closed equations. All in all, I think social behaviour curves are a simple and good tool for understanding an aspect of social reality.
There are lots of anecdotes about choosing the unused path and being the disruptor, but I feel this post explains the idea more clearly, with better analogies and boundaries.
To achieve a goal you have to build a lot of skills (deliberate practice) and apply them when it is really needed (maximum performance). Less is talked about searching for the best strategy and combination of skills. I think "deliberate play" is a good concept for this because it shows that strategy research is a small but important part of playing well.
I think this post points towards something important, which is a bit more than what the title suggests, but I have a problem describing it succinctly. :)
Computer programming is about creating abstractions, and leaky abstractions are a common enough occurrence to have their own wiki page. Most systems are hard to comprehend as a whole, and a human has to break them into parts which can be understood individually. But these are not perfect cuts, the boundaries are wobbly, and the parts "leak" into each other.
Most commonly these leaks happen because of a technical/physical simplification like forgetting that a byte overflows at 255 or electrons have travel time. However, these leaks could happen due to social simplifications too, like getTodayPosts
means "the things that get put on the top of the feed" for one and "the things which had the most engagement today" for another. Social errors are often downplayed in technical circles, which is why I think this post has an important message.
I think this post describes an important idea for political situations.
While online politics is a mind-killer, it (mostly) manages to avoid "controversial" topics and stays on a meta-level. The examples show that in group decisions the main factor is not the truth of statements but the early focus of attention. This dynamic can be used for good or bad, but it feels like it really happens a lot, and accurately describes an aspect of social reality.
If you heard "good is the enemy of the great", then also consider "perfect is the enemy of good".
Have you factored in the cost of task switching and meta-strategy research? A lot of economic theory trivializes the energy required for thinking, which might be correct for larger entities, but it's an important factor for individual humans.
Switching to a different worldview and re-evaluating your options takes a non-negligible amount of mental energy. Learning about new opportunities and hearing new strategies also takes time. If you're optimizing for doing good instead of collecting opportunities, then there could be a point where just doing your current best is better than the expected values of new ideas if you add the processing costs into them.
Thanks for the post, it's an important update on the state of information warfare.
Privacy can be thought of as a shield. If you build a wall against small-arm spam, then it's ok, but if you try to build an underground bunker, then it's weird because only Certified Good Guys have access to advanced weapons. Why are you trying to protect yourself against the Certified Good Guys?
What changed is that thanks to AI advancements in the last few years, it become possible to create homemade heat-seeking infomissiles. Suddenly, there are other arguments for building bunkers.
Have you heard that the medium is the message? It was written before the internet happened, and said that society becomes tv-like or radio-like if it watches a lot of tv or radio.
It is interesting to see how this idea applies to the internet. I agree with you on that we should not handle the internet as a block, because each side has it's artifacts. I think there should be more explaration of ideas, but on other sites. LW in it's current form suited for long essay type posts, which I think is good for it's stated purpose, methodical discussion of ideas.
It's evening, the sun is set. A man walks up to a scholar:
"Scholar, the sun rose yesterday and today morning. Will it rise again tomorrow?"
"Man, I don't know, it's kinda dark right now. Have you heard about the no free lunch theorem?"
Marketers, scammers and trolls are trying-to control the internet bottom-up, joining the ranks of the users and going against internet institutions. While it's a problem (a possibly big one), a worse situation is when the internet instutions themselves start using LLMs for top-down control. For a fictional example, see heaven banning.
That's the second filter, because "optimizing" is two words: having a goal and maximising (or minimising) it.
First, one has to aknowledge that solving aligment is a goal. Many people does not recognize that it's a problem, beacuse smart robots will learn what love means and won't hurt us.
What you talked about in your post comes after this. When someone is walking towards the goalpost of alignment, they should realize that there might be multiple routes there and they should choose the quickest one, because only winning matters.
Thank you for writing this post, I think this is a useful framing of this problem. For me personally, the doom game is fun, imho I have more motivation to do things and I become more self-confident. (if it ends what worse could happen) But that's for me, with my socially isolated Math/ComSci/CosHo background.
For others, I don't think it's a good game. I kinda noticed the tons of psychotic breakdowns around the field and, like, that's bad, but I could not have articulated why it was bad.
And even for me, I might kinda overshoot with the whole information hazard share-or-not thinking. It's better if you're in charge of the game and not let the doom game play you.
ITT: Millenials lamenting about the decadence of youth :)
On a more serious note, awesome.
I like reading text, prefer transcripts of podcasts to the actual audio (I find it boring, even when sped up), and spend too much time looking at memes on reddit and facebook. Sometimes I yell at clouds.
Somehow, videos are missing from my media consumption. I attribute it to me using 3rd party apps for everything. These apps create a barrier in the endless flow, and I have to choose content more intentionally.
I started watching the videos, and holy shoes, you found the right buttons. If your vids any indication of what's going on on these platforms, I'll update towards tiktok being actively harmful for cognition. (Not a critique to you; but the weapons you showed are symmetric and powerful, so it's possible there's enemy action there.) I can imagine myself getting addicted to those, I guess I got lucky.
All in all, I think it's a good project because I believe rational memes are good. If there's a fentanyl crisis, and you're selling heroin, then it's better to have rats on heroin.
What about weekends? There are currently 104 days in a year where you're not supposed to work.
The big difference is that these days are uniformly distributed through the year, and aren't in a one or two week block.
Political polarization is very high in the US. This is a global phenomenom, and in other countries polarization is currently decreasing.
(Ep. vibes: I went to few EA cons, and subscribed to the forum digest.)
I blame EA. They were simply too successful.
There are the following effects at play:
- Bad AI gonna kill us all :(
- Preparing for emergent threats is one of the most effective ways to help others.
- The best way to have good ideas is to have a lot of ideas; and the best way to have a lot of ideas is to have a lot of people.
- Large funnels were built for new AI Safety researchers.
- The largest discussions about the topic happened at LW and rat circles.
- The general advice I heard at EA conferences in late Feb/Mar (
notice the spike! it's March, before the big doompostedit: it's really after the doompost, I misread the graphs) is that you should go to LW for AI-specific stuff.
What a coincidence that the AI-on-LW flood and the cries for the drop in EA Forum quality happened at the same time. I think with the EA Movement growing exponentially in numbers, both sites are getting eternal septembered.
I think the solution could be to create a new frontpage for ai related discussions, like "personal blog", "LW frontpage", "AI Safety frontpage" categories. Or go through the whole subforum routes, with childboards and stuff like that.
Great story. I haven't thought you could cross steampunk and singularity, but it kind of works.
Not like it gonna matter (<100), but if it did, I don't want future me to do the funni.
Once every month. (at least as I know it)
In section 1: When the 'affine' word is commonly used, the mixture coeffecient (denoted there by ) can be any real number. When , then it's still an affine combination, but a more precise tern could be 'convex combination'.
However, as you work with function spaces, convexity in a function space is different then being a convex function, so maybe some new notation should be introduced.
Have you heard about Infra-Bayesianism?
If I get it correctly, the core idea is that "consider every possible scenario, use a maximin policy while caring about conterfactual branches", which is very similar to the idea presented in the linked post. The "Nirvana trick" in the other post is similar to just eliminating branches/cells, where the agent would take a different action from the predicted policy.
Non-Nashian Game Theory is Pareto optimal, Infra-Bayesianism implements Updateless Decision Theory. If the two are connected, that could mean that UDT and Pareto-optimality are connected too.
One of my favourite Gettier-like problems is about black holes.
Say you have a very dense star. It is so dense, that the gravitational force on its surface is capable of pulling back even the particles of its light, leaving only a black hole in the sky. How large can it be with a given mass?
It's an easy exercise using Newtonian mechanics. Take a light particle with mass . Its gravitational energy at a distance is , and its kinetic energy is at the start. If the total energy is negative, then the path of the light particles will stay within a boundary. Therefore, the answer to the question is , if the object is smaller than this, then it will be a black hole.
Of course, for that dense objects, Newtonian predictions break down. We should care about curved spacetime and use general relativity in our calculations. The answer (to my knowledge) is the Schwarzschild radius, which is .
Interesting post, abstractions are the few stable-ish quantities that weren't eaten by chaotic noise.
Exponentially growing errors are not always chaotic. Suppose you have around 1000 starting cells, with a 1% error in the population size. The number of cells doubles in each hour. The absolute error of the population size can be 10.24 times larger than the initial population 10 hours later; however, the relative error remained 1%. (The billiard ball example is still chaotic, but the tilde character does the heavy lifting: 31.4 with 10% error is an imprecise but usable measurement, sin(31.4 +- 10%) is garbage.)
If the relative error of a quantity remains bounded as the elements of a system interact, then this value could be a useful abstraction.
I do not think that the prototypical scientific method is not valuable in the long term.
In any experiment, there are lots of naturally varying parameters (current phase of the Moon, air pressure, amount of snow on the slope), and there are lots of naturally constant parameters (strength of gravity, room temperature, amount of hydroxyhypotethicol in the solution). There are base and derived parameters. The distances from the sun and the orbital periods vary between the planets, but (distance)^3/(orbital period)^2 is constant.
In the experiment, you measure X and Y. If X vary, but Y is constant, then they probably have no relation. Suppose that we want to find out that is X related to B or C. We control B to vary, and set C to a constant. If X vary, then it is not connected to C, if X is constant, then it is unrelated to B.
In the second scenario, you try to find the minimal set of base parameters that are related to X (growth rate). After some testing, we found that (growth rate) ~~ (initial age). After we found that connection, we can rule out the uncontrolled varying parameters, but there may be a connection between X and an uncontrolled constant parameter. It is possible that (growth rate) ~~ (initial age) times (1 + (amount of hydroxyhypotethicol)), and the first scenario will test these kinds of connections.
It is not enough to find which parameters won't affect the experiment. It is also important to find out which parameters could affect the experiment.
I think a useful concept would be the colour of bits. For example, a digital song can be bought on a CD or downloaded from the internet. The computer does not see a difference between them, because it just sees a number, but in the eye of the law, one of them is legal, the other is not.
The number on the CD is coloured "green", the downloaded number is coloured "red". Green numbers are legal, but red numbers are not. If you upload a song a from CD, it will be red because you can only send red numbers. However, if the studio produces a new CD, it will have green numbers because they have the copyright to the song.
Anyone can copy a digital artwork because it is just a number, but the copied number will be "yellow" coloured. With an NFT you do not buy a number, you buy the right to make this number "blue". This right can worth a lot of money if a blue number has a higher value than a yellow number.
Hi!
I am a Mathematics university student from Europe. I don't comment often, and English isn't my native language, sorry for any mistakes in my tone or my language. I'm reading this site since March, but I heard about this site a long time ago.
I was always interested in computers and AI, so I found LW and Miri in 2015. But I didn't stay at that time. I think my entry point was when someone (around 2018 maybe?) recommended Unsong on reddit because it was weird and fun. I read a lot of stories on the rational fiction subreddit. (Somehow, I did not read HPMoR. Yet.) This March, there was a national lockdown, I got bored, so I looked up again SSC and this site. Since then, I read a lot of quality essays, for which I'm thankful.
During this winter, I try to participate more in online communities. I am interested in any topic, and I know a lot about mathematics and computers, so I might write something adjacent to these subjects.
Best,
CoafOS