Savant

But once you start blaming some users for site's slow growth, you enter a slippery slope. Because if site's growth keeps being slow, then you must logically ban even more users because you start from assumption that current users are responsible for that.

We had the community vote on how to handle Debby. They voted for a permanent ban. The mods have been more lenient than that so far. Maybe the idea is that the community itself is going down the slippery slope, but I'm not sure that's true. People have different thresholds for bans, and when the median voter thinks the next ban is unreasonable, that's where the line will be drawn. There are a lot of other people some mods wanted to ban, but we haven't banned them due to the community vote.

Humans are very bad at fixing their mistakes

Well presumably you were able to identify those mistakes if you know they exist, which proves my point.

Humans can catch other human mistakes better than AI can catch other AI mistakes though.

Hmm...fair enough.

So my impression is that there are basically three groups:

Group A (Sexist people) - This group basically wants to discriminate based on sex. I think regardless of what terms we use they're gonna say things like "people with vaginas should stay home and clean things." (But in effectively whatever vocab they would normally use.) And if they were stopped from doing that they would still discriminate based on gender and say things like "anyone who identifies as a woman should stay home and clean things." They see sex and gender as connected to social roles.

Group B (Normies) - This group isn't sexist and probably only makes generalizations as far as "people with XY chromosomes are on average stronger than people with XX chromosomes." They don't really understand gender beyond that the term was at one point was commonly used to describe sex and now it means something else. They don't see gender as connected to social roles.

Group C (Trans people/allies) - Doesn't want to be told what to do on the basis of sex and uses gender to describe something inherent to identity that is hard to describe in concrete terms. And will point to gender often being used in a different way than sex historically. They see gender as connected to social roles.

I think that Group C saw Group B using terms like "man"/"women"/"gender" to describe sex when Group C was using them to describe gender, and assumed that Group B was aligned with Group A. When I think Group B and Group C were just using different sets of definitions for things. Group A can be said to be conflating the concepts of sex and gender, while Group B is more so not using the modern definition of "gender" as its defined in sociology, probably because they aren't super familiar with sociology.

From what you're saying, I thinking the issue might be that whenever Group B and Group C agree on some term to describe sex, Group A comes in and uses the term to be sexist. And then by continuing to use the term, Group B would continue to be conflated with Group A, or at least seen as enabling them.

Is this broadly accurate in your view? Obviously neither of us speaks for everyone, but this is my impression. I might be oversimplifying things.

I think I've made it clear that this matters a lot to me. As far as the quality of research I'm willing to do, the medium doesn't matter.

That makes sense. If this is a topic you're willing to discuss, then there is a question I wanted to ask (and apologies if you've addressed this before and I missed it). Has less to do with empiricism and more to do with semantics.

It's my impression that trans people typically have a strong aversion to being called "man" or "women" on the basis of sex as opposed to gender. Yet I don't sense as much of an aversion to terms like "assigned female at birth" or "assigned male at birth." Sex isn't strictly binary, but with the exception of intersex people, you can generally categorize people in to groups of mainly having female vs. male sex characteristics (to my knowledge, trans people don't really dispute this). More so pre-op vs post-op, but the point is that one could select some group of criteria they are going off of and then categorize sex based on that. It's kind of an arbitrary thing to categorize people off of, but so is the amount of melanin in one's skin, or how tall someone is, or the color of their hair. And it's still common to use words like "red-headed" or "tall" or "black" to describe people. Or even age, which like sex isn't easily observable but you usually just take someone's word for it.

So my question is: in your view, with regard to misgendering, is the aversion to sex categorizations being made in the first place, or just the terms "man" and "woman" being used? Like if people started using the terms "blargh" and "zog" instead of "man" and "woman," and blargh and zog referred to preponderance of particular sex characteristics, do you think most trans people would still be opposed to that?

I could speculate about this, but it seems like something someone more familiar with this topic would be more qualified to answer.

Voting needs to be simplified, without the need for a lengthy accompanying explanation...Time is precious.

Poorly written votes mean debaters need to do more debates in order to get an accurate elo. Which takes more time than if voters write out their reasoning in the first place. I think the current system largely accounts for this by having high standards mainly for UltraRated and requiring people to vote themselves before debating.

Also the current voting quiz/rubric hopefully makes the standards easier to understand.

Statistics: The Criminal Justice System - RAINN says that 98% of Rapists don't get prosecuted.

You mean convicted. Also that source is referring to sexual assault, not rape.

In other words, women should know that rape in America is illegal (but almost always unenforced). So (as much as I don't like saying it), treat rape as something that is legal.

That's not the same thing. Illegal acts, particularly those that harm people, carry a stigma with them that legal acts don't carry. People are generally taught from a young age not to break the law. Plus just being accused of rape carries a significant stigma. So even if we ignored morality, rape isn't really comparable to legal actions just because the law isn't always enforced.

Also, a 2% chance of 20 years in jail (avg sentence for rape) is 4.8 months in jail. So if they expect a 2% chance of conviction, a rapist should expect 4.8 months in jail on average. Not much of an incentive to rape.

Ultrarated debates need some votes. Go to your profile > badges > then take the quiz to be able to earn vote badges, then click earn vote badge to be assigned a debate to vote on.

The barrier to entry for an argument in a competitive debate that a judge takes seriously is infinitely lower than an argument in a normal discussion that a person takes seriously

This is true but it also depends on the debate format. It's not really the META here, although maybe it could work. I think on this site, most debaters would have a pretty easy time arguing against someone who advocated nuclear war so long as they did put in effort to argue against it. (In written debates, the element of surprise counts for a lot less.) Plus, even if someone wins with the nuclear war argument, it's not like people can't look at the reasonable arguments made in the debate and learn more about the issue.

And it's less about digging deep enough and more about presenting the facts as faithfully and honestly as you can, which isn't the case with a competitive debate, where there is a definitive winner and loser and you're encouraged to not make yourself the loser, regardless of what points your opponent brings up.

On the other hand, this means both sides are incentivized to make the best case possible for their side and put in a lot more effort toward research than they probably would otherwise. Most trials have lawyers who are paid to argue for one side or the other, and their goal is to win by whatever (legal, usually) means necessary. Yet this is basically the best system anyone has come up with for determining guilt or innocence, and I think it does pretty well compared to most internet arguments.

SatanLucy said it was better than 5-10 times. I was responding to Alyssa.

Or you could just, y'know, exercise some restraint and spread out 2-3 cigarettes throughout the day.

I love how smoking 3 times a day is your idea of restraint.

Of course he was, typically attacking an opposing opinion with the accusation of "destroying free speech"

Please show me where I said, or even vaguely implied, that “my opinion cannot be challenged because I say so”

SatanLucy wasn't saying you personally were destroying free speech, he was saying you were providing an excuse to destroy free speech.

Yeah yeah, and believing the Earth is not flat depends on trusting the sources saying it isn’t. Most people aren’t going to go around interviewing astronauts, The issue isn’t the primary sources and evidence, it’s the secondary sources claiming the evidence exists and is so abundant. Blah blah blah.

This is true mostly. With the Earth it's a little different from the Holocaust, because you can run experiments yourself, and it's easier than verifying a historical event. But for most people, it does matter how much they can trust secondary and tertiary sources.

If Germany legally restricts holocaust denial, it’s because in context, they have seen the violence that the antisemitic agenda can produce, ISIS recruitment is illegal in many countries, there is good reason for that too

Inciting violence is different because it's a coordination problem. Restricting speech fails to make people smarter or more knowledgeable, however it could effectively prevent them from coordinating on specific violent actions. Also I'm not buying your assumption that if a state restricts speech, they must have a justified reason for it. See: every totalitarian state in existence.

The fact that you mix “conspiracy” with the huge lie that “no discussion is being had” about one of the internet’s most virally discussed topics is absurd

I was talking about a discussion on the "reliability of the evidence" which you were saying shouldn't be had/isn't being had when Holocaust denial is banned. If there's no cross-examination, authoritative sources become less authoritative.

There is plenty of holocaust denial on the internet too, and a mountain of discussion about the evidence, and of course antisemitism is invested in denying that reality, but again, that is a matter of agenda, it has nothing whatsoever to do with the facts.

So we agree that secondary and tertiary sources can lie. That's the point. There needs to be a high cost for prominent sources to lie, and that is best accomplished through free speech.

Many in these groups see free speech as a tool that can be used to suppress dissent

If they're suppressing your speech, they're not supporting free speech, they're just lying.

despite the incessant accusations that disapproval amounts to suppression of free speech, no it doesn’t

Where did I make these "incessant accusations"?

I don’t disapprove of free speech, I disapprove of antisemitism

So we agree. I'm not sure you understand what my position is if you are challenging me yet still support free speech.

I have a right to express my opinion; you should stop trying to destroy free speech.

I don't think Lucy was saying you don't have a right to your opinion, just saying they disagreed with it.

The Holocaust is one of the most well documented and widely testified events in history

Believing this depends on trusting the sources saying it's true. Most people aren't going to go around interviewing all Holocaust survivors. The issue isn't the primary sources and evidence, it's the secondary sources claiming the evidence exists and is so abundant. I trust those sources, but only because people are allowed to challenge them and they can withstand scrutiny.

There is nothing to be learned from a discussion about the “reliability of the evidence”, the discussion unavoidably becomes “why” do you question the reliability of the evidence, in the end, the only relevant question is “what is your agenda”?

The fact that no discussion is being had makes the evidence less reliable. Claims of evidence and documentation can be fabricated. An organization could pay a dozen people to claim they are experts who thoroughly studied the issue and then lie about it. For experts to be reliable, I have to believe there is a cost to lying (i.e. that people would be able to call them out).

“Get it all on record now - get the films - get the witnesses -because somewhere down the road of history some bastard will get up and say that this never happened.” ― Dwight D. Eisenhower

There are plenty of fake quotes on the internet. This one is real, but the point is that your argument here isn't using primary evidence, it's using secondary and tertiary sources talking about the evidence. I have to trust those before I can consider the evidence, and free speech is important for that.

(Marked as offensive for obvious reasons. Using a clearly extreme example to make a point, because when people argue about free speech, it's usually about extreme examples.)

Here's an article from the grandchild of Holocaust survivors. It's in the Globe and Mail, the most widely read newspaper in Canada on weekdays and Saturdays. A company like that has a lot to lose, and so I'm inclined to trust them not to invent a genocide.

Here's a Chinese government source claiming that the oppression of Uyghur muslims is a hoax. If I were to look for dozens more sources in China, they would likely all agree, more or less. Now, these sources have to say that, because China doesn't have free speech, but the same could be said about the previous Canadian source I cited. Holocaust denial is illegal in Canada, so I should expect major sources to all report that it happened when discussing the topic whether or not it actually did.

In fact, I would hold that many sources reporting about the Holocaust in countries where Holocaust denial is illegal, have similar reliability issues to a major Chinese news source claiming that China has never committed human rights abuses, or that COVID-19 originated in the US. The government isn't strictly scripting the narrative for news sources, but banning denial is definitely putting strong pressure to write the narrative one way. (And once a government that bans speech, I'm often less inclined to believe that they aren't pulling some strings behind the scenes.)

I still believe the Holocaust happened, but mostly because I have sources from countries where it is legal to deny the Holocaust, and because the bans on Holocaust denial are poorly enforced. (Even a Chinese citizen with access to the internet and a good VPN could pretty easily outsmart government propaganda.) However, the consequence of criminalizing Holocaust denial has been that, logically speaking, many sources reporting on the Holocaust are no longer trustworthy. If the Holocaust was a hoax, the cost of lying would be low, because the government is telling them what position to take. There's other forensic evidence we can look at, and types of research individuals could do themselves, but most people won't go to that much effort, and we are generally pretty reliant on secondary and tertiary sources faithfully reporting the truth.

When we think about "knowing the truth," there's a certain type of information that comes to mind. Did the Holocaust happen? Was the Civil War fought over slavery? We often think that if people know the answers to these questions, we have effectively taught them a sufficient amount of the truth.

But there is another type of information that is arguably more important. What is the evidence that the Civil War was fought over slavery? What is the strongest argument that it wasn't? What is the strongest argument to believe the Holocaust didn't happen? This second type of information is necessary to see and analyze before we can arrive at the first type.

A logical person will be less likely to trust sources reporting a fact if the government mandates that fact. Generally, we want people to be logical, and we also want people to know the truth. If we censor falsehoods, then we can no longer achieve both of these goals. Banning Holocaust denial is killing the goose that lays the golden egg.

Most elo farmers on the previous site were contenders accepting debates from noobs who had written poor descriptions. The hope right now is that UltraRated debates will be a better measure of skill, and I would take the Rated leaderboard with a grain of salt anyway, since both instigators and contenders can game the system.

We'll have the next poll on CoC rules in a few months though, and we can always put this issue on there. Until then I wouldn't want to unilaterally go against what the current vote landed on. You're right that having the instigator set the rules up front can sometimes skew things, and there may be better ways to deal with it than what we've done so far.

Comment on the debate and tag the creator asking them to add/change a rule.

You can add rules to debates you start, or negotiate rules with the instigator and they can edit the description. So you could have a rule that says "No AI use" or that specifies what exactly it can be used for.

Site dev here.

Rules on AI should be hard established

They are, based on a community vote.

Open format should be allowed with options such as contender picking a side of the argument or having the positions randomized. I find it much more fun to debate any position as it demonstrates true ability in debating as opposed to having strong talking points on just one side.

That exists. It's called UltraRated debates. To participate in one, when you are creating a debate, set "type" to UltraRated. However, there are some requirements for those, including earning voting and topic proposal badges. Go to your profile > badges and get three of each to qualify for UltraRated.

It is AI transforming input using weights according to your goal.

The idea is you could look for weight combinations that pursue your goal and stumble upon weight combinations pursuing another goal, that appear to pursue your goal.

No, because we can see the code AI makes. We can see whole code for new AI made by AI. Because new AI doesnt use weights, all code is clearly visible, and harm easier to detect.

I don't think it would be easy to check for hidden vulnerabilities or exploits though. Plus to get a RAG that is x intelligent, you need a black box AI that is at least x intelligent, or a lot of human effort, which will fall behind companies willing to use black boxes. The black box AI just existing is dangerous because humans can be manipulated and computers can be hacked, and the technology existing means someone else may use it much less carefully.

We didnt succeed in abolishing nukes tho

We’ve succeeded in reducing the number and adding more safeguards.

Disclaimer: I did some more digging, and I think Constitutional AI is more promising than I gave it credit for. When Claude had hidden goals, it still pursued the spirit of the constitution it was given. So deceptive alignment may not be as hard a problem as I thought, at least based empirically on recent models. I don’t know how well this will generalize to smarter models though.

Weights dont have a goal.

AIs can exhibit goal-oriented behavior even without having conscious desires, and some combinations of weights absolutely lead to goal-oriented behavior, such as the ones AI has now. Through RLHF, AI learns to act as if it has the goal of helping me. So either it has the terminal goal of helping me or some other terminal goal that is aided by acting helpful; regardless, AIs are perfectly capable of exhibiting goal-oriented behavior.

[RAG] works just as efficiently as regular AI in retrieving information, tho slightly worse at reasoning. But its not like today's AI is much good at reasoning anyway

Reasoning is what companies are focused on optimizing right now. They’re not about to switch to RAG without significant pressure. If you want them to switch, then great, that’s a massive safety improvement. But if we want companies to care about safety, then everyone needs to understand the significance of this problem.

And second, we can have AI program it.

That’s just kicking the can down the road, because then we need to align the AI doing the programming.

because all countries and groups wont agree

We managed to stop nuclear war and prevent some countries from getting nukes. Pausing AI will be much more difficult, but slowing it might be feasible.

literally anyone can have a local AI, entirely programmed by themselves, trained on whatever data

Not anyone can build premier AI models. Right now, it requires extremely high compute and teams of researchers. We know who is the closest to AGI right now, and forcing them to test more or even slow down a little would reduce risk.

It is like throwing letters around and expecting to form sentence "destroy humans".

But even less likely, because not all weights control output equally.

The issue is we only judge weights by the outputs they produce. However, for every set of weights that have high capabilities and set the goal “do the task humans want,” which gives us the desired output, there are far more combinations of weights with high capabilities that say “maximize x random thing,” which is accomplished best by doing what humans want right now and playing nice, then betraying humans. So we are basically tuning weights until we get a high-capability program that does what we want right now, but most high-capability programs that do we want right now would betray humans, we just don't know which are which.

It's a bit like generating random numbers until you get a number between 1-10, but 9 of the numbers between 1 and 10 kill you. You could say “most numbers won't kill us,” but most of the ones that don’t kill us also don't do the tasks we want, so we keep tweaking the weights until we get a model that does what we want, and if the model is capable of scheming, we probably get one that kills us. When models are dumber, most models with arbitrary goals aren't clever enough to trick us, so they get filtered out. When models get really smart, even the ones with goals that require betraying humans will be smart enough to play along.

RAG AI system doesn't use any weights

Typically it’s paired with an LLM that does use weights.

AI which is RAG plus rule based logic doesn't need weights at all

Rule logic needs to be programmed manually by humans and takes far longer. Good luck getting companies to prioritize safety over performance, when they’ve shown far too little interest in safety so far.

Restricting AI development is self-defeating, because the only restricted ones will be the good guys. Evil leaders won't restrict their AI development, but will keep developing full speed.

Pausing AI development globally would require international cooperation of the likes we’ve never seen before, something akin to restricting the development of nukes. I don’t know if it’s possible to ever accomplish, but I want people to know the risks, because that’s the only way humanity has a fighting chance. Eliezer Yudkowsky, who is even more pessimistic than me, stated, “The moratorium on new large training runs needs to be indefinite and worldwide. There can be no exceptions, including for governments or militaries. If the policy starts with the U.S., then China needs to see that the U.S. is not seeking an advantage but rather trying to prevent a horrifically dangerous technology which can have no true owner and which will kill everyone in the U.S. and in China and on Earth.” That’s a tall order, and I don’t think we could ever get the kind of pause that makes Eliezer happy, but with some cooperation we might be able to slow development as a whole or require more testing and red lines. Getting everyone to understand the threat that this poses is a start, and that’s what my goal is right now. Society will decide how to handle this, but I’d rather they decide on a policy from a position of knowledge of the risks rather than a position of ignorance.

Suddenly I think of genies in lamps,

Whether one should risk a wish, not knowing whether it 'might be an evil genie who will twist your wish.

It’s worse than that, because the AI isn’t guaranteed to stay true to our wishes at all. All we can do is measure whether a particular set of weights appears to create an AI loyal to humans, and per Goodhart’s law, we should not expect that strategy to actually produce AIs loyal to humans, particularly once they are much smarter than us. We’re much more likely to stumble upon an AI with an arbitrary goal than one loyal to humans, since both will act equally helpful past a certain level of intelligence (one due to actual loyalty, the other because it’s an effective way to attain more power until it can betray us).

AI is trained to follow prompt made by humans. It doesnt follow arbitrary goals, but goals set by prompt.

We can only verify that in some use cases. The code keeps getting tweaked until the AI follows prompts in the short term. But saying that is all the AI cares about is a bit like saying, “We filtered out all the politicians who were mean to us, so these politicians are guaranteed to always act in our interests forever.” We have no idea what’s in the AI model’s head or what the politician is really thinking, we just know that their current algorithms make them appear helpful.

In future, I assume AI will follow set based logic learning like math does instead of predicting next thing in data.

Those models exist, but they are much less efficient. There’s a big gap between code simple enough for us to understand and behavior intelligent enough to do anything a human can do.

the AI you see today, "predicting next thing" patterns, is just how it is today, and it isnt superintelligent despite being trained after trillion parameters made from data.

Again, we don’t directly program in the “predict next thing” behavior, we just know the weights of the current models appear to lead to that behavior. And there are other things we want AIs to do beyond predictions, like avoiding naughty words. That’s not directly programmed in, it’s guess and check, and the “check” is very imperfect.

What I disagree about is AI being dangerous by default or in most cases.

Those randomly generated patterns aren’t dangerous by default, but to make the artificial intelligence “intelligent” you need to train it to exhibit goal-oriented behavior. So yeah if we just let it behave randomly it wouldn’t try to take over humanity, but at that point it’s not even artificial intelligence. Once it’s intelligent/trained, it has one of a billion possible goals which all appear the same to us and perform the same on tests. Most of those goals require the AI’s survival but not human survival, hence the danger.

As long as they can mark some pattern as bad, it means AI can't use it to produce output.

Sure, but there are a large number of patterns and goals that are bad for humanity (in fact, most are, because they are met more efficiently if the AI depowers humanity so we can’t interfere).

The reason why AI tends to get repetitive, why every conversation feels similar, is because so many bad patterns were removed or placed down that only a minority of patterns is producing output.

Those are the short-term ones we can identify. A smart AI could fake alignment, so we can’t generalize its behavior too far into the future.

As I said, solution is linked AI models which supervise each other.

Superintelligent AIs could collude with each other. In fact, they could probably do it easier than humans since they don’t have emotions to get in the way, and at lightning-fast speed.

AI only has as much power as you give it. Sure, you could give AI control over nukes now. Would it destroy the world? Probably at some point. But the point is, don't give AI dangerous stuff.

AI already has access to dangerous stuff—people. If it wanted to destroy humanity all it needs to do is develop an efficient virus that kills everyone and then tell a terrorist group how to make it. Heck, as long as doctors and world leaders are using it, it could probably manipulate humans into doing almost anything catastrophic if it’s patient enough (which a misaligned superintelligent AI almost certainly would be).

Humans control patterns generated by AI

They don’t control the implications of those patterns though. Let's say you can pick a random number and one of them leads to extinction, but you don't know which one does. Are you controlling the AI's behavior just because you control its weights?

as long as you dont expose AI to bad stuff, it wont learn anything bad,

Not true. AIs don’t learn to be evil from humans, they lie because it's the most efficient way to accomplish their goals. AIs will bluff in poker or be ruthless in games even when they are told to be helpful and honest, because that strategy leads to them winning the game more often. AIs learn to lie from reinforcement learning, not because lying is in their training data.

Those predictions are compared to correct answers using a loss function (function that measures how far the model’s prediction is from the correct answer)

This only tells us that the AI does what we want in the short term. A smart, Machiavellian AI will do what we want in the short term, so loss functions don’t prevent smart AIs from being Machiavellian. How an AI behaves when it is controlled by humans isn’t indicative of how it will behave when it can kill all humans.

It is still humans who determine what patterns are good or bad.

And all a human can say is “this pattern did what I want in the short term.” That does not extrapolate to how the AI will act when it has more power than all humans combined. Politicians will kiss babies and promise to help everyone, and then when they are elected they will accept bribes and embezzle money.

AI just does what its told

It does what leads to a lower loss value. That’s not the same as doing what it’s told.

This feature can be built within AI to identify evil people and refuse to follow their direction.

And what if the evil people are the ones building the AI and refuse to build that feature? Do you trust China, or Google, or OpenAI to be paragons of virtue? I certainly don’t, and even if we could, that doesn’t solve the problem of the AI itself optimizing for hidden goals.