This is a definitely an assumption that should be challenged more. However, I don't think that FOOM is remotely required for a lot of AI X-risk (or at least unprecedented catastrophic human death toll risk) scenarios. Something doesn't need to recursively self-improve to be a threat if it's given powerful enough ways to act on the world (and all signs point to us being exactly dumb enough to do that). All that's required is that we aren't able to coordinate well enough as a species to actually stop it. Either we don't detect the threat before it's too late or we aren't able to get someone to actually hit the "off" button (literally or figuratively) in time if the threat is detected. And if it only kills 90% of humans because of some error and doesn't tile its light cone in paperclips, that's still really, really bad from a human perspective.

This is definitely a spot where the comparison breaks down a bit. However, it does still hold in the human context, somewhat, and maybe that generalizes. 

I worked as a lifeguard for a number of years (even lower on the totem pole than EMTs, more limited scope of practice). I am, to put it bluntly, pretty damn smart, and could easily find optimizations, areas where I could exceed scope of practice with positive outcomes if I had access to the tools even just by reading the manuals for EMTs or paramedics or nurses. I, for example, learned how to intubate someone, and to do an emergency tracheotomy from a friend who had more training. I'm also only really inclined to follow rules to the extent they make sense to me and the enforcing authorities in question can impose meaningful consequences on me. But I essentially never went outside SOP, certainly not at work. Why? 

Well one reason was legal risk, as mentioned. If something went wrong at work and someone is (further) injured under my care, legal protection was entirely dependent on my operating within that defined scope of practice. For a smart young adult without a lot of money for lawyers that's fairly good incentive to not push boundaries too much, especially given the gravity of emergency situations and the consequences for guessing wrong even if you are smart and confident in your ability to out-do SOP. 

Second, the limits were soft enforced by access to equipment and medicine. The tools I had access to at my workplace were the ones I could officially use, and I did not have easy access to any tools or medicines which would have been outside SOP to administer (or advise someone to administer). This was deliberate. 

Third, emergency situations effectively sharply limit your context window and ability to deviate. Someone is dying in front of you, large chunks of you are likely trying to panic, especially if you haven't been put in this sort of situation before, and you need to act immediately, calmly, and correctly. What comes to mind most readily? Well, the series of if-then statements that got drilled into you during training. It's been most of a decade since my last recert, and I can still basically autopilot my way through a medical emergency based on that training. Saved a friend and coworker's life when he had a heart attack and I was the only one in the immediate area.

So how do we apply that? Well, I think the first two have obvious analogies in terms of doing things that actually impose hardware or software limits on behaviour. Obviously for a smart enough system the ability to enforce such restrictions is limited and even existing LLMs can be pushed outside of training parameters by clever prompting, but it's clear that such means can alter model behaviour to a point. Modifying the training dataset is perhaps another analogous option, and arguably a more powerful one if it can be done well, because the pathways developed at that stage will always have an impact on the outputs, no matter the restrictions or RHLF or other similar means of guiding a mostly-trained model. Not giving it tools that let it easily go outside the set scope will again work up to a point. The third one I think might be most useful. Outside of the hardest of hard takeoff scenarios, it will be difficult for any intelligence to do a great deal of damage if only given a short lifetime in which to do it, while it also is being asked to do the thing it was very carefully trained for. LLMs already effectively work this way, but this suggests that as things advance we should be more and more wary of allowing long-running potential agents with anything like run-to-run memory. This obviously greatly restricts what can be done with artificial intelligence (and has somewhat horrifying moral implications if we do instantiate sapient intelligences in that manner), but absent solving the alignment problem more completely would go a long way toward reducing the scope of possible negative outcomes.

Distinguishing between a properly cited paraphrase and taking someone's work as your own without sufficient attribution is not trivial even for people. There's a lot of grey area in terms of how closely you can mimic the original before it becomes problematic (this is largely what I've seen Rufo trying to hang the Harvard admin woman with, paraphrases that maintained a lot of the original wording which were nonetheless clearly cited, which at least to me seem like bad practice but not actually plagiarism in the sense it is generally meant) and it comes down to a judgement call in the edge cases.

A professor I know fell afoul of an automated plagiarism detector because it pinged on some of her own previous papers on the same subject, and the journal refused to reconsider. Felt very silly, like they were asking her to go through and arbitrarily change the wording she thought was best just because she had used it before because the computer said so. I think she ultimately ended up submitting to a different journal and it got accepted there.

Yeah, the joke for professors is you can work any 60-70 hours of the week you want, so long as you show up for lectures, office hours, and meetings. It's got different sorts of pressures to a corporate or industry position, but it's not low-pressure. And if you're not at the kind of university that has a big stable of TAs handling a lot of the grunt work, you're gonna have a number of late nights marking exams and papers or projects every semester, unless you exclusively give students multiple-choice questions.

Also, getting to the point of being a tenured professor is a process in and of itself. Not getting tenure means you likely get laid off.

One other thing a lot of people are missing here is that most "professors" at universities today are not tenured, or even tenure-track. They're adjuncts or sessional lecturers, who are paid more along the lines of $70k a year (often less) for what is in practice a similar workload with similar education requirements, except consisting entirely of teaching, with literal zero job security. Sessional lecturers sometimes find out only a couple of days or weeks in advance what they are being asked to teach for the semester, if anything.

Those who are good teachers will continue to be good teachers. Example of this from a prof I know who won teaching awards and continues to teach basically the same way now that she's gotten tenure. She likes teaching, that's a big part of why she's good at it, she's not about to phone it in. I think what's slipped a bit post-tenure is the amount of resources she devotes to publishing her research. I don't think her actual research has slowed down any, because she also likes that part, she's just not focused on getting papers out the door ASAP because her continued employment no longer depends on it.

There are universities with better teachers, but they tend to be those that focus on their undergrad programs, and not the big prestigious ones with massive endowments. They'll hire people who aren't as prestigious in their fields but are good at teaching (and get a nice discount on staffing costs). The prof above works at such a university; part of her job interview was giving a lecture to a room of students and faculty, and how well she did was IIRC part of the reason why she was hired in the first place. The same tiny university was a pioneer for major chunks of the current paradigm for undergrad engineering programs, and the program director spent a sabbatical at the University of Waterloo improving its undergrad engineering program. That being said, even at a more teaching-focused university like that one, there were still some real bad profs (and not necessarily people with tenure or who did a lot of good research, just inexplicably bad teachers).

One of the most frustrating things about the Blanchardian system for me is how it flattens a ton of variation into the "auto____philia" category, asserting the same "erotic target location error" cause for all of it, and the people pushing the theory tend to brush that off by asserting that trans people are lying (or grossly mistaken) about their own experiences and sexualities. That category in the original study was extremely heterogenous (and subsequent studies have been almost exclusively run on members of crossdressing fetish forums and similar, as you point out), but the sample size was too small in that initial study for any variation to rise to statistical significance. It's not simply that auto____philia is a less satisfying narrative, it simply doesn't track with many trans people's actual experiences at all, and I think analyzing a larger dataset (especially if the analysis were done by a researcher less prone to motivated reasoning and questionable statistical practices) would demonstrate that.

Blanchard would likely call me an autogynephile simply by virtue of my being bisexual, not being very stereotypically feminine, and transitioning late (though as soon as I had the knowledge and ability), despite having essentially zero erotic feelings around being feminine and never really even cross-dressing, much less getting off on it. In fact, he would likely suggest that I was mistaken or lying about the latter.

I think Phil understands the trans position (or should I say positions, because in actuality trans people have a ton of different and conflicting ideas about gender, sexuality, sex, and so on) very poorly, or at least represents it very poorly in the quote you provided. What he's presenting there is what I tend to call the "lies to cis children" version, intended to try and get the basic idea across to people who know nothing about transness and have no framework for understanding our experiences. It's also to some degree a narrative which was imposed upon trans people by doctors like Harry Benjamin and (as the concept of "gender identity") John Money (and made a prerequisite for accessing care, a strict template which for trans women included being stereotypically feminine, having zero interest in using one's natal genitals, and being exclusively attracted to men, i.e. the HSTS category) more than one trans people created. In fact, the HSTS subset of the initial Blanchard study sample likely consists almost entirely of patients who were seeing him in order to access medical transition and had to either have or pretend to have experiences and motivations which fit the Harry Benjamin template in order to do so. Most other trans people I've talked to would call that quote from Phil a dramatic oversimplification of their beliefs at best.

One of the reasons why there isn't a coherent position about gender, etc. among trans people, however, is because we're mostly just trying to get by, we're not nearly as concerned with theorizing. We're working from our own experiences, we're more concerned with practical things like access to the medical care which pretty demonstrably helps us even if we don't understand exactly what's going on under the hood, and we all have different experiences which inform how we think about this stuff. For some trans people that gender identity framework works pretty well, even if it's oversimplified. For others, it couldn't be more off-base. We're all blind people touching an elephant and trying to explain it to blind people who've barely even heard of elephants and aren't touching this one. Blind people who often prefer to ignore what we say and make up their own explanations, despite us being the ones touching the elephant. It's not easy.

I do want to pick out one thing from that explanation though, the idea of transness as broadly a developmental error to be ameliorated rather than a psychological/psychosexual condition. I actually think that, based on how we are successfully able to alleviate gender dysphoria in actual people, that categorization makes a certain amount of sense. Gender dysphoria isn't like almost anything else in the DSM in terms of how you treat it, and that treatment is dramatically more effective than the treatments available for pretty much anything else in the DSM. You basically treat it like an endocrine disorder and a birth defect, rather than a psychological condition, and that works. Treating it like the latter, or like a fetish (both of which tend toward pushing the person away from transition), doesn't result in good outcomes. 

I don't think that necessarily implies a whole lot about the underlying causes, whatever they are (I have an autoimmune disorder, celiac disease, that I treat like it's a severe allergy, these things exist), but in the absence of any real knowledge of the underlying causes, I think a "duck typing" sort of approach is a sensible one, and regardless of what the actual cause might be, it's fairly clear what works and what doesn't. You don't put someone with celiac on standard immunosuppressants or immunoglobulin and have them keep eating wheat just because it's an autoimmune disease and not an allergy. That would be a great way to harm them, not heal them. You have them cut out wheat completely and stay vigilant about cross-contamination, and they pretty much always get better. Likewise with gender dysphoria.

Yeah, that's my main issue, too. I know the original incredibly well, I worked out the chords on piano from scratch years ago. So while I get the motivation here I would really have trouble with the adapted version. 

I natively have higher expectations in terms of congregational musical and rhythmic ability, due to where I grew up (Congo), so I always feel the need to push back when people dumb down songs for group singing. My brain expects random untrained people to be able to do melody and descant, syncopation and pick-up notes, and so on, because that's what I grew up with, though I know that's not necessarily the case here, not with this demographic.

One thing that would work is to have part of the song being sung by the leader, not the whole group. That might be a workable way to incorporate the bridge back into this version of Level Up. Drop back to just low accompaniment and have the best singer do that part solo, then bring people back in. If you were to attempt something more like the original, you could also do this with the start of the song, where the beat is a lot less consistent. Have the leader start the song and then bring people in as you move toward the first chorus and the groove really kicks in.

Also, I wonder if an unclear understanding of the time signature of the original (or an attempt to fit it into something more standard) is causing issues. It's pretty much all in 7, especially once the beat gets going, and a good rhythm section that can hit the accents right really makes everything quite easy to hit in proper time. There's a quick and tight 1212123 (with the occasional 1231212) through the whole song (though for the into and first bit of the first verse it's a lot more nebulous) and most of the "challenging" notes actually land on that first beat of the 7. But yeah, you'd have to have the band really work on the song to get it to a place where you could lead it well in its original form.

My experience is that there's a huge amount of conservatism among classical instrument players. 

One part is the sound change. If something changes the sound of their instrument in a way that isn't clearly positive it's very hard to convince people to switch to it. Sometimes this is pretty rational, (I don't think anyone thinks plastic brass instruments sound better) but other times it's a kind of superstitious audiophile sort of logic. For a flute, I think bending a formerly straight tube will always have some effect on the sound. If the tube is already bent one or more times (as a bass flute must be even if the play orientation of the concert flute is maintained, because you have to have a certain length of tube which would be completely unwieldy if kept straight), a different or additional bend is lower impact. We see this with brass instruments, where you can fit a tuba into a significantly smaller volume without changing the sound significantly, so long as the bell is not shrunk too much and the mouthpiece remains the same size, simply by bending the tubing in more and different ways, so that the overall lengths of the main tube and each of the valve tubes remain the same.

The other part of it is muscle memory; people have spend thousands of hours playing an instrument with a very specific configuration. They don't want to change how they hold something even if it would be more ergonomic because it would mean a bunch of re-learning, not on their main instrument. I've encountered this myself as a pianist. If you change anything about the piano keyboard too much, it trips me up. The whole part of my brain that handles the movements for piano relies on the keys being in very specific positions relative to each other (and me) and responding to my touch in specific ways. I once botched a recital because the piano provided had too high an initial actuation pressure on the keys and I hadn't been able to do enough practice on it to adjust. I would love a piano with keys about 60% the size of the standard ones, because I have small, slender hands. But actually switching to such a piano if it were to be made would be a painful process, and getting up to speed on it would likely mean that I would lose facility with pianos that have standard-size keys.

The exceptions I've seen to this, where ergonomic innovations have been readily accepted, are triggers on trombones and 4th valves on some other brass instruments. Those have become very popular over the years. I think the reason for the difference is that a trigger or 4th valve is completely optional to use. You can still play everything exactly as you would have before, you just now have an additional option which makes certain things more convenient. There's a gradual adjustment process where you integrate the new feature into your play and you don't experience a phase where you feel hampered by the change. That makes switching costs much, much lower.

Apologies in advance for any tonal issues in this.

I have serious issues with the autogynephilia analysis, both yours and more generally. You've likely heard the critiques that many cis women qualify as AGP if they take the same surveys, but I also think the questions and scenarios are simply not useful in terms of distinguishing a sexual motivation from any other sort of gender euphoria/dysphoria/identity. Someone with entirely non-sexual motivations for transition would still likely find a sexual fantasy where they are their transitioned gender in one way or another arousing and could thus easily be a false positive on your AGP/AAP scales, in as much as those are supposed to separate out a specific type of (trans) person or reason for transitioning. Many, many entries on your list of common AGP/AAP fantasies fall into that category and can only classed as abnormal by virtue of being had by someone of the "wrong" birth sex. Given that, I really do not understand why you believe it more likely that AGP/AAP is the upstream, causal condition and not gender identity, given the inability of the tests to distinguish between the two possibilities. It seems like you're massively privileging the Blanchard/Bailey hypothesis in spite of the major issues (including arguable fraud) that you admit their research has (and I see little reason to assume Hsu's research will be any better quality, given that he clearly works extremely closely with Bailey).

Parsimony isn't everything, humans are as you say complicated, but when you're working with a typology that claims two separate, distinct causes for a transgender identity and the surveys used to analyze the supposed phenomenon can't easily distinguish between a normal variation in sexuality for someone's (identified) gender and a primarily sexual motivation for that identity despite that being an explicit goal (on top of the other issues with the research), it seems likely that you have a problem with your hypothesis. It seems particularly likely that this is the case when there is another hypothesis that matches the data at least as well, doesn't posit multiple causes for what seems like a single phenomenon, and also matches with the reported experiences of the vast majority of transgender people (where the Bailey/Blanchard typology being true would essentially require that the vast majority of trans people be lying to themselves).

And if AGP/AAP measures don't really work for their intended population, trans people, why would they be useful measures in cis people, as you're using them here?