As I heard the current discourse about scaling laws, is that we trained the models on the entire (textual) internet, and still haven't found superintelligence. They started to use video and audio data, but these sources are less meaning-dense (encoding a picture of a room takes as much data as two books of of characters), so we could be at the edge of scaling unless a new, more data-efficient model arises.
Let's assume humans can generate 40 bit/sec meaningful data. Estimates of information density in natural spoken languages converges around this number cross-culturally, so it could be a natural processing limit in the brain. Quarter-third of the day is spent sleeping, and not every interaction is captured (yet. Distopian surveillence will increse it, but) for now let's take the half of it for an avarege day. 20 bit/sec * 86400 sec/day ≈ 210 KB/day/human. Population of 8G means 1640 TB/day. A post I found with DuckDuckGo says 402M TB/day is created on the net, which would be a 0.4% mening-to-data efficiency, and given a large amount of it is video, it is beliveable. GPT-2 was released in 2020, so let's take 5 years, this would be 2.8 EB of meaning, and given a token is 16 bit, this would be 1.4E of tokens in the past five years.
Plugging it into the Chinchilla scaling it would mean an 0.00048 loss-floot-contribution, which would be insignificant compared to the ideal generative model, which has a 1.87 lossfloor. D/N=20 means 700P of paramaters, about 1000 times larger than current models. Let's assume a 3.7x increase/year, it would mean at the end of transformers will be reached in 5-6 years. But this is some back-of-the napkin calculation, I think it's more like an upper-bound.
They're in conflict with the unmeasurable unknown part of the world. As I read it (correct me if I'm wrong), if there is only a single probabilistic model (bayesian reduction?) then it will follow the expected maximum.
If there are multiple possible models, then it will assume the worst model. But to assume better, something like Kierkegaard's leap of faith needed, believeing a model not based on reason. I mean, the infra-parts tries to say something about very unknown models, so it's somewhat on the boundary of reason, and it's possible that the logical apparatus stops.
The (ancient) Greek form of debate or dialogue was based on the notion of common good. If ONE of the participants feel bad about it, then EVERYONE loses it. Yeah, during the dialogue the partner (opponent?) will look dumb, but afterwards they reach a conclusion, they learn something and part happily.
Trolling on the other hand is just a quick crack at the other's worldview. The point is provoking a response from other's, not educating and lifting them up. The motive of ending with MUTUAL respect is missing.
Like, dialogue is a martial art; you can fight with it but point is mastering one's own body. On the other hand trolling is pure optimization for K.O. punch. Pure optimization in the misalignment sense: The troller would even hurt/humiliate themselves if it would lead to a quicker victory.
There is the notion of singular truth in the world, but accepting this cuts back too. Two side goes into the debate, one side comes out, and they will have truth. It is of course your side. Does the debate optimizes for truth?
If the truth is independent of the participants (e.g. correct interpretation of quantum mechanics), then one (of course the other) side goes into the debate with a clearly losing position. By trying to defend their villanious statements, they have to lie about the world, and muddy the waters. With chance they could win, and only repeated debate could lead to a statistical conclusion.
If the truth is dependend on the participants, (e.g. Lighthaven's colorscheme is bad. We should paint the whole blue/green, let's have a debate on it.), the the truth doesn't even exists prior to the debate. An outcome will be reached, but will it become accepted as truth? The losing side can ask always ask for a remach, until concensus is reached.
For something to accepted by a concesus, a not warring dialogue is faster method. And while a dialogue has the goal of not necessery truth but a social concensus, by allowing the participants to freely abandon positions which they no longer hold, over time people will argue only defendible positions, which is the truth.
However, truth is only a byproduct, and power dynamics sometimes can overwrite it. Yet dialogue is the best I know. If you a quicker-faster-easier way to reach the truth, please let me know.
The Nash equilibrium is
The Nash equilibrium is a solution for non-cooperative games. Politics, almost by definition, is about cooperation. Cooperative games are a superset of non-cooperatives, because in the case of no communication it reduces to one.
The word "signalling" in virtue signalling is a form of communication. Democracy has many forms, the most common is representative democracy. In it, the voter does not choose a policy directly, but chooses a delegate. The job of the delegate is to learn about the world (for example economics), and make a decision based on this information. If the decision is bad for the voters, the they will be replaced in the next election. The moral hazard for delegates happens if for the delegate the cost of losing their job is lower then the cost of making a bad decision.
From the voters perspective it means that a potential delegate should show that they understand the world and make good decisions, a.k.a. they should signal their virtues. If the voter understands the world and capable of making better decisions than current delegates, than they should run for an office. The moral hazard happens, if the cost of running for office is greater than the cost of the current delegates making bad decisions.
Our democracies are not perfect: the cost of a delegate losing their job is quite low, and the cost of running for office is quite high. But the problem does not lie in mass ignorance. You awknowledge that learning about economics has costs. The same can be said about arts, literature, physics, social sciences, or ecology. It is not possible for everyone to learn all these subjects, but it is possible to select a few of our best to learn about them. The tragedy of commons would happen if everyone would need to learn about boring (for them) subjects for years.
Nevertheless, I agree with your conclusion, that in good places you are held indvidually accountable, becase it would clearly show your virtues. Writing longform posts about economics shows that you are capable of understanding the world, and given the post's contents, you think that you are able to make better decisions than the current political delegates. Have you considered becoming/are you a political activist?
P.S.: Fish reserves might be in a bad state right now, but it's not all coordination failures. International fishing agreements happen. I recommend this book: "Game Theory and Fisheries Management: Theory and Applications".
An arbitrary constant you expect to multiply by.
Usually, k is not just an arbitrary real number, but an integer, like in .
For arbitrary constants to multiply by I think (lambda, greek letter) is used.
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.
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.
Ratio of profits
Marx's theory of capital was focused on the ratio of profits, not on their mass. Attacking it by saying we have so much good under capitalism does not disprove the theory, because it is simply a statement about the mass of goods.
Marx said that under capitalism companies maximise profitrate. (Not profit!) Suppose there are two strategies:
Strategy A will win, because it will give 1.5 dollars for each dollar of investment, while strategy B only nets 1 dollar for each dollar. The mass of profits is higher in the B case, but the focus is on return of investment.
Usually, automatization and technology improvements rises profitrate and revenue at the same time, and the life of the avarege worker improves.
Yet, independent of the total mass of revenue, the question about the ratio of distribution remains. If the ratio of wages increase, the ratio of profit would decrease, therefore the owners should want to lower the wage-ratio, while workers should want to increase it. And this is the key moment in the class struggle.