Well, I didn’t mean to propose an argument.
My impression is that there is not a convincing roadmap. I certainly haven’t seen one. However, I recognize that there is a healthy possibility that there is one, and I just haven’t seen it.
Which is why I’m asking for the white paper / textbook chapter that presumably has convinced everyone that we can expect AGI in the coming decades. I would be very grateful for anyone who could provide it.
Obviously, AGI is feasible (more than could be said for things like nanotech or quantum computing). However, it’s feasible in the sense that a rocket ship was feasible in the time of the ancient Greeks. Obviously a lot of knowledge was necessary to get from Greeks to Armstrong.
Right now all we have are DL models that are basically kernel machines which convert noise into text, right? My intuition is that there is no path from that to AGI, and that AGI would need to come from some sort of dynamic system, and that we’re nowhere near creating such. I would like to be proven wrong though!
matejsuchy
Can you refer me to a textbook or paper written by the AGI crowd which establishes how we get from GPT-n to an actual AGI? I am very skeptical of AI safety but want to give it a second hearing.
China’s sciences are not very good, and relatedly most of those papers are likely of extremely low quality. I know Chinese, and it’s a wonderful language, but I wouldn’t recommend learning it for that purpose. My 2c
I’m having trouble finding it. It was a survey done by David Putrino, it’s mentioned here:
”By contrast, Putrino told me that in his survey of 1,400 long-haulers, two-thirds of those who have had antibody tests got negative results, even though their symptoms were consistent with COVID-19.”
https://www.theatlantic.com/health/archive/2020/08/long-haulers-covid-19-recognition-support-groups-symptoms/615382/
Here is a more vague claim that seems to corroborate:“Whereas some “long haulers” were found to be positive for SARS-CoV-2 RNA by RT-PCR at symptom onset, many did not fulfill the criteria for testing at the beginning of the pandemic, or tested negative at a time when respiratory symptoms had subsided. In addition, some “long haulers” did not have detectable antibodies to SARS-CoV-2 when the first serological test (Abbott) became available commercially.
https://onlinelibrary.wiley.com/doi/epdf/10.1002/acn3.51350If you look at the demographics of the sufferers, it’s also somewhat suspect. Middle-aged well-to-do white women are massively over-represented. If it were a matter of over-active immunity you’d expect young women, and if it were a matter of a weak immunity getting overrun, you’d expect men. So, I’m not sure there’s a good reason to suspect this population really would be the main sufferers of some legitimate long-term syndrome. OTOH, middle-aged white women have sky-high rates of depression.
As for “long covid” itself, my sense from talking with GPs is that it’s mostly misattributed. There’s the notorious study which showed that 2⁄3 of “long covid sufferers” had never been infected with C19 to begin with. It seems like it’s just somewhat stronger-than-usual depression? All the risk factors for “long covid” seem to just be risk factors for depression.
On the matter of vaccine effectiveness, do we know what the numbers are for obese vs non-obese? Vaccines commonly don’t work (well) for the obese, and given how overweight America is I wonder if this is depressing our numbers. Maybe it’s like 98% for thin, 70% for overweight, 40% for morbidly obese or something like that?
Bullshit jobs are a lot of it. I’d add anything in media, a lot of academia (shocking numbers of those who’d be better off running a plumbing business in some depts), and non-profit / political activism stuff
Yes, it certainly is too short. I fear some of my writing is too long-winded, and wanted to try the whole “most blog posts should be a tweet” thing. Evidently, this is not the most effective strategy.
I don’t intend to use education as interchangeable for sanity. Here “sane ideology” is just a cultural belief that maximizes utility. The three ideologies here are: “education is not worth it”, “pursue education according to your ability”, “get a masters as long as you’re not brain dead,” which are espoused by much of the lower, middle, and upper classes, respectively.
My claim is that the middle position is the best — that wealthy and dumb individuals would actually be much happier if they learned a skilled trade and built a respectable small business than if they went to university and ended up in a low-pay, low-impact jobs program for the overeducated such as … well, I don’t want to be rude and name the sectors, but I’m sure you can fill in the gaps. Of course, it’s easier to argue that eschewing education altogether (the low-slack-enforced lower class ultimatum) is also undesirable. So, the middle class approach to education is sane insofar as it produces the best results.
Of course, insane is not quite an apt descriptor for the two tail beliefs here. However, there are other areas where the insane/sane dichotomy really is appropriate (I hesitate to introduce them, as, naturally, they are the sort of thing that induces half of the population to froth at the mouth).
In retrospect, this may be the fatal flaw in attempting to introduce something like this. In order to properly provide evidence for an insane/sane split, one would necessarily have to pick an example that is culture-wars radioactive. It’s my understanding that doing so is against site policy/norms, so I’m not sure how to approach this.
The Sanity Waterline Bulges at the Middle
Ok so I did some reading and my sense is that obligate homosexuality is not very common in the type of matriarchal hunter-gatherer societies you mention (and is not found in wild animals), but is found in domesticated humans and animals. There does appear to be some genetic component as there is a bit of heritability. The obvious question is if there is some selection effect present in domestic environments not present in the wild.
There are two hypotheses which seem somewhat plausible; in both, the gene persists largely due to low mate choice on behalf of domesticated females. In the first, (https://link.springer.com/article/10.1007/s40750-016-0056-6), domesticated females are coercively mated with males without regard to sexual preferences (by parents who arrange marriages in the human case, and human breeders in the animal case), and so the defect is not subject to much selection pressure and gains prominence through drift. At first glance, this seems more convincing for the animal case and less so for the human case — I imagine that in many arranged marriages where there is an obligate homosexual male partner, many of the children are not his. In the second, females have to compete for high-status mates in a highly stratified society (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4534200/), and the gay gene makes them especially seductive. This seems somewhat just-so to me (is a wealthier husband worth a gay brother when it comes to fitness?), but the preliminary regressions at least don’t count against it.
I like the logic of a conditional response gene, but it doesn’t seem to bear out, at least in our society. Many homosexuals manifest gender-atypical behavior well before puberty, which seems to be evidence against the hypothesis.
Also (and this is really just from-the-seat-of-pants-musing), coalitions seem to be very important for tribal social dynamics. In my anecdotal experience, homosexuals seem to have higher coalition-building skills than heterosexuals on average. If anything, they might be more threatening to the “alpha” male (to the extent one would even exist) — again, anecdotally, macho types seem to be pretty bad at playing that sort of soft-power coalition-building game.
“Societies without homophobia” — let me know if you find one! I don’t say this gleefully.
Do you believe it? An obligate homosexual sibling would need to help their siblings have an additional 4 children who survive to reproduction in order to break even. That is a significant burden — especially given infant mortality rates in the ancestral environment, we’re potentially talking about 8 additional pregnancies, at which point it seems implausible.
Ockham’s razor might tell us that LG, where the individual has a mind which motivates against reproduction, are simply the consequence of some developmental failure?
Her Internal
The hour I first alieved
Thanks for offering this insight! Could you clarify how those things are selected for in training? I am actually struggling to imagine how they could be selected for in a BUD/S context — so sharing would be helpful!
Also, you say that the training had effects but “not to that magnitude … not necessarily even in that direction.” I’m confused — it sounds like your friend enjoyed effects both to that magnitude and in that direction. Am I misunderstanding?
Also, if he did enjoy such effects as you describe, do you have any hypotheses for the mechanism? Given that such radical changes are quite rare naturally, we’d expect there to be something at play here right?
Thank you for offering feedback! The study you mentioned also references another that may indicate that further studies could be helpful to determine whether there is an effect “The results of McDonald, et al. (1988) suggest, inconclusively, that some personality changes may occur during SEAL training” (p 12). Generally speaking, your criticism is well-taken; I agree that the SEAL example is a difficult one because of the strong selection effects. Generally speaking, one should a priori expect more composite conscientiousness in any elite group (except maybe among artists?). One would have to diff-in-diff things to empirically determine an added training-based effect.
My main qualm with the selection argument is it might elide differences between sub-traits of conscientiousness. “The average SEAL is also more persistent, reliable, and scrupulous, viewing life as a series of task- oriented challenges,” whereas BUD/S seems only to select for one of these sub-traits: persistence. Given that there is status associated with being a SEAL, we might expect Berkson’s Paradox to actually lead SEALS to be somewhat lower in the unselected sub-traits (task-orientation, reliable, scrupulous). Since this is not the case, we might adopt a null hypothesis wherein we expect that there is some add-on effect from training itself.
That being said, you’re right that this is very far from conclusive. I suspect it would only really be compelling to those who personally witnessed the rapid shift in personality consequent to elite military training in an acquaintance. I count myself among this group, but recognize it may not exactly be a large % of the LW community — hopefully there the convergent cultural evolution argument holds a bit more weight?
Of course mind uploading would work hypothetically. The question is, how much of the mind must be uploaded? A directed graph and an update rule? Or an atomic-level simulation of the entire human body? The same principle applies to evolutionary algorithms, reinforcement learning (not the DL sort imo tho, it’s a dead end), etc. I actually don’t think it would be impossible to at least get a decent lower bound on the complexity needed by each of these approaches. Do the AI safety people do anything like this? That would be a paper I’d like to read.
I don’t know whether to respond to the “Once you know how to do it, you’ve done it” bit. Should I claim that this is not the case in other fields? Or will AI be “different”? What is the standard under which this statement could be falsified?