@ControlsFreak's banner p

ControlsFreak


				

				

				
5 followers   follows 0 users  
joined 2022 October 02 23:23:48 UTC

				

User ID: 1422

ControlsFreak


				
				
				

				
5 followers   follows 0 users   joined 2022 October 02 23:23:48 UTC

					

No bio...


					

User ID: 1422

I think you've lost your own thread. Rather than tying yourself up in knots and running yourself in circles, just be clear.

due to the definition of knowledge work in the first place

You don't have one of these, either.

Assembling Google, as a glorified directory, is not knowledge work, even if you consider the slightly more complex algorithms that go into its modern iterations. Full stop.

The Russel conjugation lives. Sneerers in the other direction call LLMs things like "a blurry JPEG of the internet". These sneers are not helpful, and I notice that you've now abandoned any speak of a test in terms of economic value.

"LLMs display jagged intelligence" or "spiky intelligence" or something of that nature.

Does Google Web Search display jagged/spiky intelligence? Clearly, it is not generally applicable and has major failure states and such.

Anyways, no real test needed

Then yeah, probably no response really needed from me to you. You don't seem to have much other than vibes. That which can be asserted without justification and all.

A lawyer for example is not merely an information processing algorithm

I'll just pause here for a second to observe yet another moment of how terminology is typically used very poorly in these debates. Typically, when people are trying to tout how 'intelligent' (whatever that means) LLMs are and how they're totally going to replace all humans... especially 'white collar work' or 'knowledge work' or whatever... they portray human (knowledge) work as mere information processing. An input-output process. With contextual understanding, sure. Even Google Web Search has some contextual understanding in it; it's much more complicated than just page rank, as you know. Getting back to the point, the reason why all these humans are going to get replaced is because their 'intelligence' is just merely an information processing algorithm.

It is through this slight of hand that many people live in their Russel conjugation. The things I prefer are intelligent; the things you prefer are merely information processing algorithms; the things he prefers are nothing but simple algorithms like page rank/OLS.

You seem to not quite be doing exactly that with humans, but you still haven't given me any real test to distinguish.

if enough people pay money for [knowledge] work, to AI...

This is the spot where the terminology is overloaded. It's sneaking in something about AI simply being called "AI". Why can't we replace this with a more generic term, "If enough people pay money for [knowledge] work, to an information processing algorithm..."? And thus, Google Web Search would again become intelligent.

I think one would have to argue that there is something fundamentally different, other than the name, between different types of information processing algorithms.

why don't they just pull out guns if they want to win so bad?

I just remembered that, as bizarre as it seems to somehow fold this into some sort of test of intelligence, we also have things like this. Bollocks if I know what that means about criteria for intelligence.

Page rank

Reducing Google Web Search to Page Rank is like reducing LLMs to OLS. Yes, OLS is in there, but it's a much more complicated information processing algorithm than just that.

Fundamentally, the point is that no one has a definition of 'intelligence' that is any good. Your test wasn't just that it produced value. Your test was:

The billions of dollars generated by LLMs come from them performing tasks that, until very recently, could only be done by educated human minds. That is the fundamental difference. The value is derived from the processing and generation of complex information, not from being a physical commodity.

I responded to your test, but you seem to not have responded at all to my response to your test.

Hmm.. I suppose, in the interest of fairness, we need to exclude the skills of human chess GMs too. After all, they've trained extensively on chess data. Lotta games played, and openings memorized.

I mean, I don't think so? But how would we know? What test would we use to distinguish?

Very little ability to extrapolate outside the training distribution

This seems not entirely true.

why don't they just pull out guns if they want to win so bad?

Whereas this just seems bizarre.

How exactly do you think learning works?

I mean, do you really want me to give a full explanation of the entire field of ML? There are many different varieties. [EDIT: Do you think that all algorithms that use 'learning' are "intelligent"... or just some of them? How do you know the difference?]

If you think just learning from existing data is illegitimate

That's not really what I said. I just said that one thing that we can conclude from the premises you presented was that a bunch of chess was in the training set. You had wanted to conclude instead that it meant something about intelligence. I sort of don't see how... primarily, because I don't think almost anyone has a justifiable definition of intelligence that allows us to make such distinctions from such premises.

The billions of dollars generated by LLMs come from them performing tasks that, until very recently, could only be done by educated human minds. That is the fundamental difference. The value is derived from the processing and generation of complex information, not from being a physical commodity.

Prior to LLMs, would you have said that Google Web Search was intelligent? Prior to Google Web Search, it likely took an educated human mind to figure out how to find answers to all sorts of complex information problems. It generated billions of dollars in value by processing and generating complex information. Sure, it sometimes sucked... but LLMs sometimes suck, too.

The fact that an LLM can even play chess, understand the request, try to follow the rules, and then also write you a sonnet about the game, summarize the history of chess, and translate the rules into Swahili demonstrates a generality of intelligence that the Atari program completely lacks.

I mean, no? It just means that there was a bunch of information about chess in its training set.

@Lizzardspawn for visibility

I think there is a 'basic' answer here, just in terms of energetics.

Looking at a single atom version of the reaction they're talking about to create gold, the idea is that you have an atom of mercury-1981, and you hit it with a neutron. If all goes well, the end result is one atom of mercury-1972 and two neutrons coming out. You can just compute the energetics of this reaction just from mass/energy conservation. This is, indeed, one of the first things I computed when going through the paper. The answer is that the reaction is endothermic, which is similar to what you might have seen in chemistry - the reaction requires you to put energy in in order for it to happen. The way you typically put energy in is to have a fast-moving neutron that is flying in to hit the mercury-198 atom. When you do the calculation, the required energy in for the neutron is just under 8.5MeV. You must have a neutron flying at least this fast into a mercury-198 atom to accomplish the desired reaction.

Common uranium-235 fission reactors do produce neutrons flying around; that's necessary for them to keep the chain reaction going. But the energy of those neutrons is low in comparison. It does produce a spectrum of neutron energies, but the peak of that spectrum (the most number of neutrons produced) is around 0.7MeV, the average being about 1.9MeV (it's a bit skewed)3. You can find that the spectrum does continue to tail off toward the higher energies, but eyeballing the chart, you have about a two-and-a-half order of magnitude reduction in the production of neutrons that are at the sufficient >8.5MeV range than you have at the lower energies. If I actually integrated the curve, the number of sufficiently energetic neutrons produced would surely be <<1% of the total neutrons produced, and the question really is about the number of zeros I should put after the decimal point before we get something non-zero.

Now, fast breeder reactors. They do also split plutonium, which does produce a slightly faster neutron spectrum... but it's not much. The curves are quite close to U235. The 'fast' part of the name is just that they use the (primarily ~1-2MeV) neutrons they have directly (when they're "fast", where "fast" means >1MeV) rather than slowing them wayyyy down with a moderator like they do in traditional reactors.

That is, the short answer is that existing fission reactors just don't produce enough neutrons that have enough energy to convert mercury isotopes (see footnote 2 again). Whereas with deuterium-tritium fusion reactors, the primary reaction is just H2+H3=>He4+neutron. If you do the energetics here, assuming worst case scenario with no kinetic energy coming from the input hydrogen atoms, you still get neutrons coming out with just over 14MeV. That's plenty of energy to hit some mercury and get what you want. If, of course, you can design your reactor right (and there are a bunch of other considerations that I won't get into here; just this basic consideration of energetics should be sufficient for the instant question).

1 - Mercury-198 is a 'relatively' abundant natural isotope, about ten percent of all the naturally-existing mercury in the world. In gathering it up, you'll likely be digging it out of the ground. Then, you take that ore and process it until it's the kind of stuff you want. Generally, people use chemical/physical means to get rid of other stuff and 'isolate' the 'good stuff'. This is the 'enrichment' bit.

2 - Thereafter, Mercury-197 naturally decays to gold with a half-life of like 20-70 hours (I didn't go back and look up the exact numbers for this comment).

3 - A weirdness that requires getting into looking at cross-sections is that they actually prefer even slower neutrons for further U235 fission. This is why they have 'moderators' in reactors - to slow neutrons down to a speed that is best for further fission events. There's fun back story here in the history of the development of ideas for the possibility of sustained fission; it took some work to figure out which isotopes of uranium would split with different energies of incoming neutrons; it turned out to be important that U235 would do fine with slower neutrons (which it could, itself, readily produce), whereas U238 required faster neutrons and couldn't sustain itself with its own production of neutrons.

This was a really fun paper to read, especially since I just noted that I'm going through an MIT OCW nuclear course right now. My actual knowledge on the topic still rounds to approximately zero, but it was actually enjoyable to just go through the proposed reactions/decays, just pull up the same tables they're using, do the incredibly simple energetics calculations, and see that they are, indeed, correct. I would have had no clue how to do even that just a few months ago.

So, can confirm that the stone simple energetics work; they're not so far out to lunch that they've made such a stupidly basic error (we're not dealing with total cranks). I can't say much of anything on any of the many many other questions involved concerning reactor/process design, materials handling, economics of it, etc. They do point out some prior works that had looked into this in the past, so it's also not unprecedented, but the current authors get an order of magnitude more production in their calculations. The current authors, correctly in my view, point out that the prior works (in the 80s) didn't really show their work for how they got their estimate for gold production, as they were focused on cobalt (and the current authors write reasonably significantly on mercury enrichment, which prior works didn't, and I don't have the knowledge to evaluate). There may be (and probably is?) some other technical barrier to the rest of the scheme that an experienced nuclear engineer would spot in an instant, but if not...

What a time to be alive!

I've thought about this for a long time, but have been dwelling on it more lately. You are absolutely right that no one bothers to teach kids to understand stuff. Back when I was teaching in university, many many experiences (often via my own attempts to teach the students to understand stuff) made that abundantly clear.

However, I am somewhat sympathetic to their plight. In many cases, you don't know how far along you can take the student. Often times, you have a known end point to a program, at which point, the vast majority of your students will simply not continue further. So you have to pick your battles.

Like, for example, I get why the intro calc classes for undergrad engineers do what they do. You've only got them for so long; many of them really only need so much; you have other things to get to. Of course, for me, it all seemed so absurd in retrospect. Why couldn't you have just started me off with Baby Rubin? It's really not that hard of an on-ramp; it begins with friggin' sequences! But it does take some time and growth, and let's be honest, the vast majority of the engineering students who go through the college will stop with a bachelors, and when you think of all the types, the chemical engineers, civil engineers, hell, the industrial engineers, etc., they'll probably get by without really understanding.

I never did any nuclear stuff, so believe it or not, I'm going through an MIT OCW nuclear course right now. I haven't "taken a course" in a long time. Wowza is there a lot of stuff in there that makes me want to say, "Ya know, if you had just had your students take a full set of quantum classes already, you could have actually done this right, in a way such that they could really understand what's going on, rather than being a bit handwavy and saying 'this is just what happens because of quantum magic that we don't know yet'." ...but how long would the curriculum take to get there? These kids have basically just taken ordinary differential equations! I honestly kinda wonder to what extent they get what proportion of their undergrads to really grok it within the four years, or if they still have plenty of clean-up to do in grad school.

As much as I often hate on the unis, I am sympathetic to their constraints here. I don't know what to really do to fix much of it, because I do think one of the roots may be the utterly disastrous K-12 situation from a long legacy of terrible public control with the primary mission of babysitting and only secondarily happening to have any learning going on (perhaps due in part to their own, must-take-all-comers constraint).

infinite resources

Supply curves slope upwards.

the core being something like "doing unpleasant or boring yet necessary things, in a timely manner without prompting"

Thanks for the other context above, too. Combined with this, I'm not really sure this is really a matter of willpower/executive function/conscientiousness. It seems a bit more like just the human condition that some things are unpleasant, and different folks make different tradeoffs. I think there are a variety of reasons why folks make different tradeoffs here, too.

Whence your urge to find solutions to problems that actually work? When it comes, how does it manifest? [my question]

I'm not quite sure I've worded this in a way to get at what I'm looking for. The way you made it sound originally, it's like you have some excitement or something for certain types of problems. I was kind of getting Lottery of Fascinations vibes, and wanted to see if it was that sorta thing. I'm not really sure your response gets at that, because I probably didn't word the question well. Things like deferring to guidelines doesn't really seem to fit, unless the idea is that you sometimes rabbit hole down an alternative interpretation that hit the lottery. Or personally, do you just get tweaked by some types of problems or whatever.

I think, big picture, and going back to your earlier comment, you mention that you've lost more weight with diet/exercise in the past. That doesn't really seem to me to be describing someone who just lacks willpower, especially given the assumption (that I am not solidly making) that willpower is the necessary ingredient for such a thing.

Instead, it feels to me to be more of just the typical things of choices and tradeoffs. Like, you said that you'll plan on going hypertrophy über alles post-semaglutide, but hypertrophy sucks, man. I mean, yeah, some people enjoy dreamer bulking (which never really accomplishes the dream) on the diet side, but you still have to lift lots of heavy weights a bunch. Seems a bit weird, since you say that you just don't like the gym. But then, I guess you maybe say you'll do it because you feel like you kind of have to, but that gets back to feeling like a willpower-limited thing again (in the model where willpower is the necessary ingredient). So I don't know. Is there something that makes 'hypertrophy über alles' trip your fascination for finding solutions to problems (because yeah, those exist, and they involve lifting a lot of heavy weights), but that fixing your diet doesn't? Perhaps you could introspect some more on some of the differences and see if you could find relatable components to the problems, see if you could trick your lottery into finding at least something relevant in there to grab your attention.

Maybe I'll start with one little thing. Hypertrophy doesn't work great with a terrible, junky, dreamer bulky diet. I mean, it works okay, but if you're fascinated with the idea of hypertrophy über alles, you might want to consider the problems that it poses... problems which do happen to have solutions which actually work.

My family, as much as they love me, still regularly sigh and tell me they wish I was less lazy or had more willpower.

In what contexts?

Firstly, why would I lie about my willpower?

I would not claim that you are lying.

willpower... executive function... conscientiousness

I think these are probably nonidentical concepts, though I would likely have to spend some additional time thinking about it to be able to write on it eloquently.

Whence your urge to find solutions to problems that actually work? When it comes, how does it manifest?

I find this and the discussion below rather fascinating. It's pretty clear from multiple responses, including some of your own, that you don't simply lack willpower. And it's not at all like some folks would have you believe these conversations go down, where there's a bunch of folks (made of straw or something) telling you that you just lack willpower or are a stupid failure or something. Instead, you're trying to self-proclaim a lack of willpower, which is mostly contradicted by all available evidence.

And further, instead, you've not described almost any challenges that in any way really resemble any sort of lack of willpower. Most of the actual challenges you've described are just problems with a variety of known solutions that actually work... and, well, you've also proclaimed that you have an urge to find those sorts of things.

Frankly, as I put it:

There are a bunch of reasons why they don't do it, and that's okay.

I don't think you've quite hit the nail on the head yet for why you don't do it, but I think it's pretty clear that it's not a matter of willpower, and it's probably not really a matter of a couple minor challenges that have a variety of pretty well known solutions, either.

So, in this sentence, what is the "problem" that is in need of a solution? Is it, like, "the problem of trying to decide what to tell people"? Or what?

Then by all means, please elaborate on how you intended the following sentence to be interpreted:

I'm a doctor for many reasons, but ranking highly among them is that I have an urge to find solutions to problems that actually work.

Ah, the good old "all that's needed is just some willpower" argument.

Ah, the good old strawman. Actually, very very bad old strawman.

Compare what I wrote:

There are a bunch of reasons why they don't do it, and that's okay.

I don't know to what extent a clustering can be identified that can be simply labeled "not wanting to put in the effort".

So educate me, then. Because the phrase 'Changing your lifestyle does actually work; it's just that many people don't do it.' falls pretty well in line with what my attitude would have been a year or more ago.

You went wrong a single sentence later:

Like you, I felt the majority of weight-gain and weight-loss issues was a matter of people simply not wanting to put in the effort.

'lifestyle changes can sometimes only work to a point'

Possibly so. I'd need to see some high quality research on this question to know much either way, where those points might be, whether they can be predicted, etc.

Like you, I was of a similar attitude.

I don't think you have accurately captured my attitude. In fact, I think you have gotten it completely wrong.

And as time has gone by, I'm becoming more and more convinced that our modern diet has done extreme damage to our bodies

Perhaps so. Biological processes in general do not seem to be fully-reversible, especially when you include the effects of aging. Nevertheless, that is not an argument against the measurable physiological benefits of certain lifestyle changes.

Oh, well then this is just the standard, all-too-common, strawman. You're responding to a figment of your imagination, not anything I've written.

What I'm saying is that advising lifestyle changes rarely works. I don't have firm figures at hand, but I suspect that the number of people I've recommended such eminently sensible things like losing weight, stopping smoking and going to gym grossly outweigh (pun not intended) the number who actually did anything about it.

This is much more circumspect than your original comment. The problem of advice-giving is significantly different in nature, and it has significant dependency on a wide variety of external, contingent factors that are not-necessarily related to the typical time-independent, mechanistic processes that the biological and medical sciences study. If you would please kindly continue to be this circumspect in future comments, that would be appreciated.

Maybe "But don't tell people that utterly destroying enemy doesn't work, because it does." would be better?

I don't know how this analogy is supposed to work. The point of the development of military doctrine is to build up a body of professional knowledge, generally to the purpose of, indeed, destroying the enemy (though there are sometimes tweaks for political constraints or other political objectives). This is, indeed, intended to be "what works".

Is the point of your analogy that the endeavor of developing military doctrine is simply fallacious from the get-go? This has other implications that I can think of. For example, rather than moving TRADOC, as Trump did, I think this point of view says that he should have simply eliminated it altogether. Of course, I think you can tell that I don't think that this is the point of the analogy, but I'm kind of struggling to see what the point is.

Maybe, on the other end, it's something along the lines of, "It's not terribly helpful to be a 400lb guy in a bed who just writes somewhere on the internet, 'Hurr durr, have you tried killing the bad guys?'"? I mean, sure? Yeah, I just don't get what you're going for, and I don't get how it's relevant to what I've said.

trying to not overeat vs extreme marketing of hyperpalatable foods is an adversarial process

Mathematically speaking, I would distinguish the two. This may be a complicated and difficult environment, but it is not an adversarial one. There are deep mathematical differences between the two.

telling them that they are stupid failures

Sorry, what? You're just off the mark. Aside from the inherent differences between adversarial processes and other dynamic processes.

I am very happy that these drugs helped your mother. I do not disparage anything about these drugs or anyone who chooses to take them.

...however.

I have an urge to find solutions to problems that actually work

You, like many others, go too far. Changing your lifestyle does actually work; it's just that many people don't do it. There are a bunch of reasons why they don't do it, and that's okay. They may be perfectly fine using a drug. Nothing wrong with that. But don't tell people that changing their lifestyle doesn't work, because it does.

Let's take something like, I don't know, becoming a doctor. I've heard that this process sucks. I've heard that plenty of folks burn out or fail at some point. I'm sure someone's mother somewhere failed in trying to become a doctor, regardless of how much her family tried to make her do it. Nevertheless, I think there are still fine reasons to say, "Here are the objective things you need to accomplish to become a doctor, and here are a variety of subjective tips to help you pattern your life in a way that is conducive to achieving that goal, if you so choose." Some people won't do it, and that's okay (in fact, the vast majority of people right now don't become doctors). We don't have a pill yet that magically gives people all the required knowledge of a doctor. But even if we did, it wouldn't be a reason to say that the other (true, good) information "doesn't work".