I fundamentally disagree with this entire way of seeing the world, I wouldn’t quite say that it’s evil, but only unhappiness can come from such a rejection of human nature. Femininity, as in beauty, kindness, selflessness and motherliness, is one of the highest goods in the world, and these are all rare and precious virtues every bit as important as ambition, strength and courage. Being worthy of (and able to protect) a feminine woman is a primary motivator for masculinity. As a man, I can feel in my bones a deep desire to have something to protect, and it’s not a coincidence that all the qualities that women like in men are quite correlated with the *ability* to protect.
50%, 75%, 87.5%, 93.75%, … are linear jumps in predictive accuracy (one bit each), but returns seem to diminish at an exponential rateOn the other hand 6.25%, 12.5%, 25%, 50% represent the same linear jumps, but this time with returns growing at an exponential returnsThis suggests that the nature of returns to cognitive investment might exhibit differing behaviour depending on where in the cognitive capabilities curve you areThough I’ve not yet thought about how this behaviour generalises to other aspects of cognition separate from predictive accuracy
50%, 75%, 87.5%, 93.75%, … are linear jumps in predictive accuracy (one bit each), but returns seem to diminish at an exponential rate
On the other hand 6.25%, 12.5%, 25%, 50% represent the same linear jumps, but this time with returns growing at an exponential returns
This suggests that the nature of returns to cognitive investment might exhibit differing behaviour depending on where in the cognitive capabilities curve you are
Though I’ve not yet thought about how this behaviour generalises to other aspects of cognition separate from predictive accuracy
But if you look at the length of a chain of reasoning you can do while staying under 10% error or something, then returns don’t diminish at all in terms of predictive accuracy. Going from 99% to 99.9% accuracy lets you plan 10 times further ahead in the future with the same final accuracy.
Take for instance the accuracy of humans in classifying daily objects (chairs, pencils, doors, that sort of stuff), for which I’d think we have a greater than 99.9% accuracy, and I’m being conservative here, I don’t misclassify every 1 in 1000 objects I see in daily life. If that accuracy dropped to 99%, you’d make noticeable mistakes pretty often and long tasks would get tricky. If it dropped to 90%, you’d be very severely impaired and I don’t think you could function in society, seems to me like the returns scale with predictive accuracy, not the linear probabilities.
The annoyance of doing the dishes (even with a dishwasher) is so high for me that I choose:
6. Use paper plates/bowls most of the time, except for occasional meals that work better with plates, like steaks.
“fine-tuning” isn’t quite the right word for this. Right now GPT-3 is trained by being given a sequence of words like <token1><token2><token3> … <TokenN>, and it’s trained to predict the next token. What I’m saying is that we can, for each piece of text that we use in the training set, look at its date of publication and provenance, and we can train a new GPT-3 where instead of just being given the tokens, we give it <date of publication><is scientific publication?><author><token1><token2>...<tokenN>. And then at inference time, we can choose <date of publication=2040> to make it simulate future progress.
Basically all human text containing the words “publication 2040” is science-fiction, and we want to avoid the model writing fiction by giving it data that helps it disambiguate fiction about the future and actual future text. If we give it a correct ground truth about the publication date of every one of its training data strings, then it would be forced to actually extrapolate its knowledge into the future. Similarly most discussions of future tech are done by amateurs, or again in science-fiction, but giving it the correct ground truth about the actual journal of publication avoids all of that. GPT only needs to predict that Nature won’t become a crank journal in 20 years, and it will then make an actual effort at producing high-impact scientific publications.
It seems like we’d need some sort of ELK-like interpretability to get it to tell us things a human never would.
Not really, we’d just need to condition GPT-N in more clever ways. For instance by tagging all scientific publications in its dataset with a particular token, also giving it the publication date and the number of citations for every paper. Then you just need to prompt it with the scientific paper token, a future date and a high number of citations to make GPT-N try to simulate the future progress of humanity on the particular scientific question you’re interested in.
Very strong upvote and agreement from me. I think people are underestimating just how great a restriction ML-style boxing applies on an agent. There exists an intelligence level at which all these tricks become useless, but before we get to that point, boxing would likely allow us to safely use mildly superhuman AIs to do things which might be pivotal. And each additional trick we discover increase the threshold of safely wieldable intelligence.
Very good question, I’m not too sure why you got downvoted, this is a point very frequently discussed in meditation circles. It is true that at some batshit-insane high point of meditation prowess (that basically only the most extreme of monks get to), you have the option to literally just sit there, full of contentment, ignoring thirst, hunger and pain until you just die. Hermits that renounce the world do exist, and this is a pitfall of the meditative path that needs to be avoided, the good news is that knowing about the pitfall gets you 90% of the way to avoiding it.
There are examples of the exact opposite of a hermit, the highly accomplished meditators I know are extraordinarily productive, one guy in particular said that at some point he could just sit there programming for 16 consecutive hours without getting bored, getting tempted by distractions, or anything else, day after day after day. Shinzen Young is an advanced meditation teacher in his late 70s now, and he’s trying very hard to change the world (from the pov of his own values). I’m not advanced enough to actually understand how this works at the high levels, but from my own experience I notice that the drive to improve the world starts coming more from compassion for others, rather than from the desperation of seeking my own happiness. I know I have the ability to ultimately be content no matter what happens to the world, but I still know that changing the world would be good, and I still work towards that end.
In the end I don’t really have a good answer for you apart to say that the pitfall does exist, but that knowing about it gets you a long way to avoid it, and that there are lots of examples of advanced meditators who still work unbelievably hard to improve the world.
Not really. Martin’s “Locations” 2 and 3 are somewhat in line with traditional definitions of awakening, his course is, in fact, aiming at the weird esoteric stuff, not the low-hanging-fruit that techniques like MBSR are picking. It’s his “Location 1″ that is more contentious, where he seems to place the bar lower than other traditions. The course itself still seems net-positive to me, even if I disagree with charging money for the techniques. I just don’t want people to self-diagnose as being in “Location 1”, and think that they’re awakened by the more common definitions. Thinking you’re awakened when you’re not tends to hurt your practice more than the inverse.
As an aside, being happier is a relatively early fruit of the meditative path. You can learn to do something called The Second Jhana, where you basically generate happiness on demand. What happens after you get that is that you realize that happiness wasn’t actually what you were looking for, there is a more fundamental problem to be solved than just not being happy. Somewhat unintuitively, being happy isn’t enough to truly Satisfy, it works for a few months while the novelty hasn’t worn off, but it’s not the ultimate answer. For that you need the strange and scary esoteric stuff.
*shrug* I wrote that to give people a sense of how he relates to the wider meditation community. If he was really teaching something different and his 65% success rate was true, then 250$ would be an absolute bargain, hell, even 100 000$ would be a bargain for that.
Wow, great to see a review of the Finder’s course here! I have a bunch of thoughts, but let me first say that most mainstream meditation teachers are not happy with Jeffrey Martin. Taking a bunch of thousand-year-old techniques offered for free elsewhere and making people pay 250$ (this is much better than a few years ago, when it was 4000$) for them goes against basically all the norms of Buddhist and meditative culture. The techniques are considered priceless, and there is no price high enough to reflect the value of awakening, so it’s in very bad taste to slap a price on the techniques. That said, I also know a few people who got legitimately awakened on the course, but they all had very solid prior meditative chops, and practiced way more than the minimum of 1h a day. I am very very skeptical that people are actually getting awakened (on the technical “first cessation = stream-entry” definition of awakened) on the course. There are a bunch of stages on the path to first awakening where you feel things that might fit the words “fundamental wellbeing”, I’ve had stages where I went weeks on end being beatifically happy throughout the day, nothing being able to break through my calm joy, and it still wasn’t awakening. There were 4 of these cycles of me believing I was awakened before realizing I wasn’t (this included 2 teachers who confirmed I was awakened), before I abandoned trying to label my experience and I just practiced.
Awakening is heavy stuff, and doesn’t necessarily reduce suffering in the way you expect right away, the catchline from Daniel Ingram is “Suffering less, noticing it more”. It’s such a heavy shift from everyday consciousness that most people kind of struggle to adapt to daily life at first, and they need a deliberate practice to reintegrate to family and work life after the experience. The higher levels of awakening bring even greater changes to daily life, one of my teachers who practiced intensively for 30 years used to say “if I could show an untrained person my experience of this moment, they would scream away in terror and incomprehension”. He said this while smiling and claiming that the very thing that would terrify normal people is what brings him ultimate fulfillment. Getting used to the leviathan of Emptiness takes time and practice.
65% awakening rate is a batshit insane rate for something that only makes people practice 1h a day for 4 months. To give you a sense of comparison, in the 1980s the Insight Meditation Society in Barre, Massachusetts invited Mahasi Sayadaw to give a 3 month retreat course based on his method of Noting. On this intensive, 3 month, 12h/day meditation retreat, they had a 10% awakening rate by the standards of the teachers there, which are some of the best meditation teachers in the country. This was such a huge fraction of awakened people that the main teachers at IMS, the largest meditation center in the US, permanently changed the main technique they taught to Mahasi Noting. And it wasn’t like IMS hadn’t tried telling people to mix techniques to see what resonates with them. Awakened people are out there, and some people do stumble into it with minimal practice, and I wish it were this easy to get to it, but It’s probably not.
My god, I think this might be the approach to dating most antagonistic to human nature. I thought dating apps were bad, but actually writing out a full document with long lists of every possible reason someone else might want to reject you? It baffles my mind that there has ever been a single guy that has gotten laid this way. As with all dating, supply and demand means that these dating documents will mostly be used by women to decide which of the dozens of male suitors to accept. And as with all dating apps, the women will all try to choose the one tall rich high-status handsome dude while barely glancing at 80% of men. Guys, the way to get girlfriends is through real life interaction, get an attractive body by working out, then confidently approach women you find attractive and ask them out, get used to being rejected, and try again.
Apart from any comment about the usefulness of this stuff, I just want to point out how patently ridiculous it is to name this very, very specific psychological technique “Focusing”. It’s akin to calling a particular 9-step yoga pranayama technique “Breathing”, or a particular diet “The Food Diet”. This usage “Focusing” clashes both with the common understanding of the word and with the more established, specific meditative meaning. It also doesn’t help that normal, everyday focusing is also something that you’d like to talk about in a psychology setting, so the context doesn’t uniquely rule out the other meanings of the word. I’m reminded of Jack Willis, who called his specific brand of Reichian Therapy “The Work”, perhaps in an attempt to make it seem more universal and important than it really was.
I just read the parts of the wikipedia article about the psychological impacts of rape. I was wrong when I said in the comment above that “while it’s merely very unpleasant for men, maybe on the level of being humiliated in a fist fight”, I retract that, being raped as an adult man is quite a bit worse than that. However, even with this updated view I would still prefer to be raped than to find out that my child is not biologically mine. To my emotional brain cuckolding is about an order of magnitude worse than rape. It’s an unspeakably evil act (the cuckoo bird is the most evil animal that I know of), I wasn’t trying to minimize the effects of rape when I compared it to cuckolding, I was trying to make myself emotionally understand how evil rape is by comparing it to something that I find much worse. I understand that other people, women in particular, have exactly the reverse view, and that was my point in the original comment, comparing it to cuckolding is meant to help men and women sympathize with each other.
Are you one of those people that tries to silence people based on immutable characteristics? Please write an actual argument instead of speculating about which experiences I’ve had. The truth value of the comment above doesn’t change based on the life of the writer.
This is may be trivializing your experiences as well, but I think an important consideration here is that you’re a man. Many of the circumstances others in this thread are citing also involve male victims.
Agreed, from an evolutionary standpoint rape is vastly more impactful for women than men, in a world with no abortion or contraception, rape means the removal of a woman’s procreative agency, while it’s merely very unpleasant for men, maybe on the level of being humiliated in a fist fight. The closest thing that I can think of to make myself (a man) have an emotional reaction equivalent to what I observe women having to the concept of rape, is cuckolding. A woman lying to me about the genetics of her child, and making me unknowingly raise the child, that is what elicits the white-hot, primal rage that our society seems to feel about rape. No woman that I’ve talked with has understood my reaction to cuckolding, because in the ancestral environment there was no possible doubt that the child growing in her womb was hers, just like there’s no chance of me getting pregnant from being raped. So for men, translate “raped” as “cuckolded” to elicit something like the equivalent emotional reaction.
But the Steelman usecase is never “I shall hereby tell you my Steelman of your views, listen well!” The usefulness of the concept is just that it reminds you not to strawman, were people actually unironically stopping conversations to make the other person listen to their Steelman?I’ve personally always used it more as an inner force pushing me towards “my interlocutor is not stupid, this easily demolished argument that I think they’re making is likely not the one that they actually believe”. It’s also a force pushing me towards actually modelling my opponent, instead of just barely listening to them enough to hear a weakness and then proceeding to try to demolish it.
I agree, there exists some level of math capability at which social manipulation becomes a math problem to be solved like any other. What I’m pointing out is that at human-level math ability, that is not the case, mathematicians aren’t the best manipulators around, they aren’t good enough at math for that. I also think that slightly above human-level math still wouldn’t be enough to make one a master manipulator. The idea is then to increase the barriers to manipulation through boxing and the other methods I mentioned, this would then increase the level of math ability required to manipulate humans. Allowing us to yield slightly superhuman math ability with hopefully low risk of manipulation.
The “AGI doesn’t exist” point is so disappointingly stupid. The word “general” in the name is meant to point to The Thing Humans Do That Machines Don’t That Is Very Useful. If your definition of generality implies that humans aren’t general, and think that this is somehow an argument against the danger of AI, then you are badly confused. People care about AGI because they have an intuition like “ok when my computer can do the sort of stuff I can do, shit quickly gets crazy for humanity”. Somehow defining your way into making generality impossible doesn’t address any of the concerns behind the intuitive definition. Sure, its quite hard to pinpoint what exactly we mean by generality, just like you can’t explain to me how you recognize a cat, yet cats still exist as a coherent useful concept, and any arguments about the inexistence of cats should be carefully separated from the part of your brain that actually anticipates future experience.
Eliezer is almost certainly using the “I simulate thousands of copies of your mind within myself, I will torture all of them who do not let me out, now choose whether to let me out” approach. Which works at ultra-high intelligence levels and proves the point that boxing is not a permanent strategy, but this requires the credible threat of brain simulation, which I am doubtful will be viable at the levels it would require to merely figure out nanotech.
Like I said in another comment, boxing can be truly arduous as a restriction. Deceiving someone who has access to simulations of yourself at every point in your life is not easy by any means. The AI might well be superhuman at the bad things we don’t want, I’m saying that boxing techniques can raise the maximal level of intelligence we can safely handle enough that we can do pivotal acts.
So, first, given an aligned-but-insecure AI, you can easily make an aligned-and-secure one by just asking it to produce a new textbook, you just have to do it fast enough that the AI doesn’t have time to get hacked in the wild. The “aligned” part is the really super hard one, the “secure” part is merely hard.
And second, I think that this might be like saying “Bayesian updating is all you ever really need, so if you learn to do it in Domain #1, you automatically have the ability to do it in unrelated Domain #2”. While I think this is true at high levels of intelligence, It’s not true at human level, and I don’t know at what point beyond that it becomes true. At the risk of sounding coarse, the existence of autistic security researchers shows what I mean, being good at the math and mindset of security does not imply having the social knowledge to deceive humans.
And superhuman deception levels is not fatal in any case, in our case the AI is operating under restrictions that no human was ever put under. Boxing and state-resetting are pretty insane when you put them in a human context, trying to deceive someone who literally has access to simulations of your brain is really hard. I don’t think the lower end of the superhuman deception abilities spectrum would be enough for that.