Exponential curves happen when a quantity's growth rate is a linear function of its own value. In practice they're all going to be logistic, but you can ignore that as long as you're far away from the cap of whatever factor limits growth.
So what are the things that could cause "AI growth" (for some suitable definition of it) to be correlated with AI?
The plausible ones I see are:
- growing AI capabilities spur additional AI capex
- AI could be used to develop better AIs
The first one rings true, but is most definitely hitting the limit since US capex into the sector definitely cannot grow 100-fold (and probably cannot grow 4-fold either).
The second one is, to my knowledge, not really a thing.
So unless AI can start improving itself or there is a self-feeding mechanism that I have missed, we're near the logistic fun phase.
> it's just mimicking the human ideas of good/bad that are tied to most "things" in the training data.
Most definitely. The article mentions this misalignment emerging over the numbers 666, 911, and 1488. Those integers have nothing inherently evil about them.
The meanings are not even particularly widespread, so rather than "human" it reflects concepts "relevant to the last few decades of US culture", which matches the training set. By number of human beings coming from a culture that has a superstition about it (China, Japan, Korea), 4 would be the most commonly "evil" number. Even that is a minority of humanity.
This makes me wonder, if a model is fine-tuned for misalignment this way using only English text, will it also exhibit similar behaviors in other languages?
It's sadly an example of terrible leading question bias, to the point where I'm surprised that it even got a 22% oppose rate.
The percentages would change dramatically were one to write it as, "From everything you have seen and heard, do you support or oppose the recent rules requiring adults to upload their id or a face photo before accessing any website that allows user to user interaction?"
Both questions are factually accurate, but omit crucial aspects.
I live in a country where 91.78% of the population voted for a referendum that bought back hard labour in prisons.
As one of the few who voted against it I have yet to encounter a single person who voted for it who both supports hard labour and realised that was in the question being asked.
Why do you claim the 1999 referendum reintroduced hard labor in NZ prisons? I've never seen anything to that effect. The reforms were related to bail, victims rights and parole.
"Should there be a reform of our justice system placing greater emphasis on the needs of victims, providing restitution and compensation for them and imposing minimum sentences and hard labour for all serious violent offenses?"
Now let's play tldr with the law!
Luckily it was non binding and stands forever as an argument against binding referendums.
I'm not really seeing the deception here since it specifies hard labour and says it would apply to all serious violent offenses. How could you vote for this and not know you were voting for hard labour?
"Should there be a reform of our justice system" -> "should the law be passed"
"emphasis", "restitution", "compensation" -> too hard to skim, brain is bailing out
---
the only way to provide valid direct democracy is to provide more than enough explanations and rewordings from both sides of the debate *at the point of voting* to remove miscommunication
The deception is that it combines two largely unrelated questions into one vote - leading with one that most will agree and followed by one that is more questionable. By the time people will be reading the second question they will already have be primed with an opinion on the first.
In many respects I agree with you there, I almost went with softer language. The fact remains that it appears people were deceived. All of the advocacy pushing the referendum only focused on the first part. To this day I find people who are amazed that it mentioned hard labour and and that they voted for it.
[edit]
I guess think of it in terms of a vote that you had discussed and decided upon before you voted. Could you honestly say that you would read every word of the question or would you just look at the start of it to establish that it was the question under discussion and then trust that the discussion accurately represented what the question on the form would say. The length of the question, was I believe specifically designed to be long to prevent the frequency of its full publication.
Could you honestly say that you would read every word of the question
Yes?? It's not like a school exam where the questions are secret until you see it in the voting booth, and even if it were, you should still read the question carefully. I'm all for things being written as clearly as possible but at some point you have to acknowledge that voters have a responsibility to think about what they're voting for.
It is consistent with my experience that most people seem to not realise that they voted for hard labour.
That is indeed the entire theme of this thread, That people can give an answer to a question that in some way does not reflect their honestly held opinion.
> most people seem to not realise that they voted for hard labour
This is incredibly anecdotal, a major victim of selection bias, and also there are possibly effects of agreeableness here b/c it seems like you may be part of a vocal minority on this issue (and I mean that with absolutely no negative connotations). That said, I don't automatically reject vibes based determinations like this because often the high bandwidth of personal interaction can outweigh the problems with low bandwidth questioning in polls. But in this case, when 90% voted in favor, I have a hard time believing it. I think that what you can safely conclude from your experience is that a lot of people didn't know what they were voting for. If you wanted to say maybe it was really 75-25 I could go with that, but 91% in favor (in an actual vote and not a poll) is pretty convincing to me.
Eh, it’s kind of the opposite for me. I’ve never seen any legitimate vote in a democracy > 90%. Even if you put ‘we agree that puppies are cute and fluffy and deserve all the pets’, > 10% will vote the other way purely out contrarian ness. Or because they’re cat people. Or because fuck you, that’s why.
And there is no way you can convince me 91% of New Zealand voters, where this is the common policy stance [https://www.mbie.govt.nz/business-and-employment/employment-...], had any clue they were voting for forced hard labor for prisoners. Especially considering how relatively cushy the current standards are for prisoners.
I’m sure with enough lawyers and PR folks could also write (and pass) a CA popular thingy which calls for all males to be kicked in the groin too.
That said, I’m also a big believer in voters getting what they voted for - only way they’ll learn. Besides, a few kicks to the groin might teach them a lesson!
Modern slavery legislation passed in 2022 has abslutely no bearing on public opinion on crime and punishment for violent offenders in 1999. People in NZ have been fed up with soft on crime policies and short setences for violent repeat offenders for a long, long, long time (and continue to be today). Despite what the noisy left wing in this country might tell you.
It baffles me that you people think we didn't know what we voted for in a referendum question expressed in a single sentence which included the words,
> Should there be a reform of our justice system [...] imposing minimum sentences and hard labour for all serious violent offences?
idk, maybe they're actually in favor of hard labour (which was after all spelled out in the question) and they're just telling you what they think you want to hear so you don't bug them about it. A lot of people are happy to lie this way.
I don't buy that, and even if they did that doesn't make it deceptive. I'm not arguing in favor of this increased punishment, it just seems to me that its stated plainly enough you can't seriously argue that people were tricked.
It is somewhat deceptive, or at least misleading, to bundle up the concepts of giving the victims compensation, and punishing the prisoners more aggressively.
Unless the prison labor is providing the compensation, but that would be totally bizarre and dystopian, haha. Not really the sort of thing you’d see in a civilized country.
"Hard labour for all serious violent offenses" seems almost refreshingly straightforward. Was there more in the actual referendum that was hidden? I grant that "serious violent offenses" is somewhat vague; was it overly broad?
That question clearly says hard labour. I'm sure some people didn't read it, but I think there also may be another effect there, where when talking to people in person, they realize you are morally opposed to forced hard labour, and don't want to seem like a bad person, so they pretend they didn't know. Sort of similar to the recent effect in the US where trump significantly underpolled as many voted for him but don't want to admit it.
Yeah. It's the "foot in the door technique." The same is being done with Chat Control.
It's very difficult to oppose a law ostensibly designed to fight CSAM. But once the law passes, it'll be easily expanded to other things like scanning messages to prevent terrorism.
See also:
> Concern over mass migration is terrorist ideology, says Prevent
The problem is that one of the most secure places in the world is a maximum security prison. Hence many measures that drag us closer to the prison state do genuinely improve security.
It takes some balls for the society to say: No, we don't agree to yield an essential liberty in exchange to actual real increase of security. Yes, we accept that sometimes bad people will do evil things, because the only way to prevent that would inflict even more damage on everyone. Yes, we are willing to risk harm to stay free.
There is always plenty of people who are ready to buy more comfort in exchange for limitations of liberty that, as they think, will not affect them, because they are honest, got nothing to hide, always follow the majority... until it does affect them, but it's too late.
> It's very difficult to oppose a law ostensibly designed to fight CSAM. But once the law passes, it'll be easily expanded to other things like scanning messages to prevent terrorism.
Oh, look, you did it in literally two sentences. It turns out it's pretty easy to to oppose such law. Only there's simply no need to do it when you're the main beneficiary.
Regarding [1], the study itself mentions that stopping watching porn reverses the effect. In layman's terms: watch enough of it and the novelty wears off, but the sexual drive returns. Hardly a harm, it's what happens with every human activity.
[2] makes the big logic jump of assuming that someone who watches kinky porn fails to separate between fantasy and reality. It is the same line of reasoning as the disproven "videogames cause violence" paradigm and it is pushed by the same sort of people (personal hypothesis: they might be projecting). This could ironically point to a problem limited to at least some individuals failing to differentiate the two, but studies find that at the population level, a higher availability of porn correlates with lower rates of sexual assault. My personal reading is that it provides a safe outlet for sexual frustration and moderate desensitization reduces the chance that someone will, so to speak, get aroused over an exposed ankle.
On [3]... you're linking to a single data point, not a series nor a correlation; additionally, even if the correlation actually existed held, people's propension to form stable relationships is a preference, not a harm. It is also not related to minors, and it is not something that the state has any business sanctioning, much less with incarceration.
I moved to Sweden three months ago. The article is good in pointing out that a cashless society has runaway effects to other social systems.
Despite being in one of the best situations possible (EU citizen, works in tech, has assistance from locals) and being proactive about it, I am still without a Swedish bank account. In order to have an unrestricted one, you need to have Swedish ID, which is a multi-step process that requires registration with a Swedish address. Having acquired ID recently, my application has been sent, but the bank can potentially refuse it.
This locks me out of BankId, which locks me out of large sectors of society. Bank accounts in my native country are supposed to be closed once one no longer resides there, but doing that is simply not an option if I want to survive.
It is very easy to accidentally design systems that lock out foreigners moving into a country. The EU must oversee member states on this aspect or this will effectively destroy our freedom of movement.
So what are the things that could cause "AI growth" (for some suitable definition of it) to be correlated with AI? The plausible ones I see are: - growing AI capabilities spur additional AI capex - AI could be used to develop better AIs
The first one rings true, but is most definitely hitting the limit since US capex into the sector definitely cannot grow 100-fold (and probably cannot grow 4-fold either).
The second one is, to my knowledge, not really a thing.
So unless AI can start improving itself or there is a self-feeding mechanism that I have missed, we're near the logistic fun phase.