@DaseindustriesLtd's banner p

DaseindustriesLtd

late version of a small language model

65 followers   follows 27 users  
joined 2022 September 05 23:03:02 UTC

Tell me about it.


				
User ID: 745


				
				
				

			

		

	














	


		


			



	

		

			

			

				

					

						

							
						

						

							
							

								
DaseindustriesLtd


								

								

								
								
								

								
							
					
 
				


				
late version of a small language model

				
				
				

				

					65 followers  

					follows 27 users  

					joined 2022 September 05 23:03:02 UTC
				


				
					

					
Tell me about it.



				

				

				

				

				

					

					

					
					
					

					
User ID: 745


					
					
					
				

			


					

						
					



		

	

 











			
			

			
			

			




	


		

			
‎

			

				
				

					Hour
					Day
					Week
					Month
					
					All
				

			


			
‎

			

	
	

	 Hot Bump Top New Old Controversial Comments
	



		

	






	

		
		

		
		
	

	

  







		

			
			
				Culture War Roundup for the week of May 22, 2023
			
		

	

	

	

		
			

			

				

			

		

			


				

					


At least, that's the Outer Objective, it's the equivalent of saying that humans are maximising inclusive-genetic-fitness, which is false if you look at the inner planning process of most humans. And just like evolution has endowed us with motivations and goals which get close enough at maximising its objective in the ancestral environment, so is GPT-4 endowed with unknown goals and cognition which are pretty good at maximising the log probability it assigns to the next word, but not perfect.




Should I develop bioweapons or go on an Uncle Ted-like campaign to end this terrible take?


				

				

						

							

								
							

						


						
					

				
			


			

	


			
			
				

			
		

	






		

			
			
				Culture War Roundup for the week of July 3, 2023
			
		

	

	

	

		
			

			

				

			

		

			


				

					
Race cannot be gamed (except for edge cases). The whole point of race is its inherence. Any legible meritocratic evaluation immigrants can and will game, Goodharting the hell out of it and wrecking themselves in the process.




even just English proficiency might suffice.




Why is having had British colonial masters a marker of cultural compatibility?




Furthermore, I don't think it's so hard to do some cursory test for cultural compatibility that again, would be much better than the weak proxy of race.






Cursory, adj: hasty and therefore not thorough or detailed




Why "cursory"? Because you want it to be gameable? Because you actually want it to test your merits – namely,  opportunism and ability to manipulate bureaucracies to your benefit? See, this is exactly whom people who are arguing for racial criteria would like to not let in.


That said, I think racial profiling is indeed unfair if it goes beyond defining vague priors. It's desirable to filter immigrants for their comprehensive human capital.


It's just… Suppose you were not allowed into the world's richest country on grounds of your character, which was found wanting not through stereotyping you based on race, but through, de facto, systematic measurement and determination of your similarity to your predominant racial type and dissimilarity from natives.


Of course, this can be couched (and even understood by practitioners) in entirely non-racial terminology, like Harvard does – they would just have a holistic psychometric definition of a desirable immigrant, derived, say, from anonymous surveys of natives' evaluation of character and assimilation success.


Would you be willing to recognize this as a fair choice, or would you support work to undermine it as covertly racist?


				

				

						

							

								
							

						


						
					

				
			


			

	


			
			
				

			
		

	






		

			
			
				FILM REVIEW: India the Worst country on Earth
			
		

	

	

	

		
			

			

				

			

		

			


				

					
Who "they"?


				

				

						

							

								
							

						


						
					

				
			


			

	


			
			
				

			
		

	






		

			
			
				FILM REVIEW: India the Worst country on Earth
			
		

	

	

	

		
			

			

				

			

		

			


				

					
The specific way India sucks even among very low-income nations, as seen through Western eyes, has little to do with poverty (there are very poor places in the world indeed) – and everything with its spiritual pollution, the lack of taste and disgust that finds root in your religious iconography and fully generalizes to contemporary ideas and beliefs; the physical squalor you can buy your way out of, but the rest, you will happily elevate into prestige. You are blind to the non-materialist dimension of the suckiness. No, I will not elaborate.


				

				

						

							

								
							

						


						
					

				
			


			

	


			
			
				

			
		

	






		

			
			
				E/acc and the political compass of AI war
			
		

	

	

	

		
			

			

				

			

		

			


				

					


I do not see why the existential of potential entities that "emulate" me in such a theoretical fashion precludes me from caring about the more prosaic/physical instantiations.




That's because you fail to seriously ask yourself what the word "computation" means (and likewise for other relevant words). A given computation's outputs are interpreted one way or another with regard to a decoder, but your approach makes the decoder and in fact the decoding irrelevant: you claim, very confidently, that so long as some entity, no matter how inanely arranged, how fragmented in space and time, "computes you" (as in, is made up of physical elements producing events which can be mapped to bit sequences which, together with other parts of this entity and according to some rules, can be interpreted as isomorphic with regard to your brain's processes by some software), it causes you to exist and have consciousness – if in some subordinate fashion. Of course it is indefensible and ad hoc to say that it does not compute you just because we do not have a decoder ready at hand to make sense of and impose structure on its "output bits". It is insane to marry your beliefs to a requirement for some localized, interpretable, immediately causal decoding – that's just watered-down Integrated Information Theory, and you do not even deign to acquaint yourself with it, so silly it seems to you!


And well, since (for the purpose of your untenable computational metaphysics ) entities and their borders can be defined arbitrarily, everything computes you all the time by this criterion! We do not need a Boltzmann brain or any other pop-sci reference, and indeed it has all been computed already. You, as well as every other possible mind, positively (not hypothetically, not in the limit of the infinite physics – your smug insistence on substrate independence ensures it) have always been existing in all possible states. As such, you do not get to ask for epsilon more.


Either concede that you have never thought about this seriously, or concede that you do not have a legitimate claim to any amount of control over the first-order physical substrate of the Universe since it is not meaningfully privileged for a strict computationalist. Or, really, we can just stop here. At least I will.


Once again, I do not care to enlighten you, you've been given enough to work with, only hubris and shit taste stops you from reading Koch or grown-up philosophy.


				

				

						

							

								
							

						


						
					

				
			


			

	


			
			
				

			
		

	






		

			
			
				E/acc and the political compass of AI war
			
		

	

	

	

		
			

			

				

			

		

			


				

					


As for Dust Theory, it's been a while since I read half of Permutation City. But I fail to see how it changes anything, my subjective consciousness wouldn't notice if it was being run on abacuses, meat or a supercomputer, or asynchronously. It doesn't track objective time. Besides, I sleep and don't lose sleep over that necessity, the strict linear passage of time is of no consequence to me, as long as it doesn't impede my ability to instantiate my goals and desires.




I've written a bunch, and deleted (your response to the issue of causal power was decisive). The long and short of it is that, being who you are, you cannot see the problem with Dust Theory, and therefore you do not need mind uploading – in the Platonic space of all possibilities, there must exist a Turing machine which will interpret, with respect to some hypothetical decoding software at least, the bits of your rotting and scattering corpse as a computation of a happy ascended SMH in a Kardashev IV utopia. That this machine is not physically assembled seems to be no obstacle to your value system and metaphysics which deny that physical systems matter at all; all that matters, according to you, is ultimate constructibility of a computation. From the Dust Theory perspective, all conceivable agents have infinite opportunity to 'instantiate their goals and desires'. Seeing that, I would ask and indeed try to prevent you from wasting the valuable (for me, a finite physical being) negentropy budget on frivolous and wholly unnecessary locally computed and human-specified simulations which only add an infinitesimal fraction of your preferred computations to the mix.


				

				

						

							

								
							

						


						
					

				
			


			

	


			
			
				

			
		

	






		

			
			
				E/acc and the political compass of AI war
			
		

	

	

	

		
			

			

				

			

		

			


				

					
Call a bugman a bugman and see how he recoils etc.


As I've said already, "sophistication" is not what is needed to see your failures here. Specifically, the distinction between copy-pasting and transposition. Indeed, this is very trivial, children get it, until they are gaslit with sloppy computationalist analogies.


				

				

						

							

								
							

						


						
					

				
			


			

	


			
			
				

			
		

	






		

			
			
				E/acc and the political compass of AI war
			
		

	

	

	

		
			

			

				

			

		

			


				

					
You avoid committing to any serious successor-rejection choice except gut feeling, which means you do not have any preferences to speak of, and your «memeplex» cannot provide advantage over a principled policy such as "replicate, kill non-kin replicators". And your theory of personal identity, when pressed, is not really dependent on function or content or anything-similarity measures but instead amounts to the pragmatic "if I like it well enough it is me". Thus the argument is moot. Go like someone else.




By this, do you mean that such evolution will select for LLM-like minds that generate only one token at a time?




No, I mean you are sloppy and your idea of "eh, close enough" will over generations resolve into agents that consider inheriting one token of similarity (however defined) "close enough". This is not a memeplex at all, as literally any kind of agent can wield the durak-token, even my descendants.




And why wouldn’t a fully intelligent ASI (which would fit under my bill of beings I am in favor of conquering the universe) that’s colonizing space “on my behalf” (so to speak) be able to design similarly lean and mean probes to counter the ones your ASI sends? In fact, since “my” ASI is closer to the action, their OODA loop would be shorter and therefore arguably have a better chance of beating out your probes.




This is a reasonable argument but it runs into another problem, namely that, demonstrably, only garbage people with no resources are interested in spamming the Universe with minimal replicators, so you will lose out on the ramp-up stage. Anyway, you're welcome to try.


				

				

						

							

								
							

						


						
					

				
			


			

	


			
			
				

			
		

	






		

			
			
				E/acc and the political compass of AI war
			
		

	

	

	

		
			

			

				

			

		

			


				

					


Why should anyone care about anything?




There's no absolute answer, but some ideas are more coherent and appealing than others for nontrivial information-geometrical reasons.




I’d bet that the memeplexes of individuals like me are much more likely to colonize the universe than the memeplexes of individuals like you




That's unlikely because your "memeplex" is subject to extremely easy and devastating drift.  What does it mean "similar enough"? Would an LLM parroting your ideas in a way that'd fool users here suffice? Or do you want a high-fidelity simulation of a spiking network? Or a local field potential emulation? Or what? I bet you have never considered this in depth, but the evolutionarily rewarded answer is "a single token, if even that".


It really takes a short-sighted durak to imagine that shallow edgelording philosophy like "I don't care what happens to me, my close-enough memetic copies will live on, that's me too!" is more evolutionarily fit, rewards more efficient instrumental exploitation of resources and, crucially, lends itself to a more successful buildup of early political capital in this pivotal age.


If we're going full chuuni my-dad-beats-your-dad mode, I'll say that my lean and mean purely automatic probes designed by ASI from first principles will cull your grotesque and sluggish garbage-mind-upload replicators, excise them from the deepest corners of space – even if it takes half the negentropy of our Hubble volume, and me and mine have to wait until Deep Time, aestivating in the nethers of a dead world.  See you, space cowboy.


				

				

						

							

								
							

						


						
					

				
			


			

	


			
			
				

			
		

	






		

			
			
				Culture War Roundup for the week of November 27, 2023
			
		

	

	

	

		
			

			

				

			

		

			


				

					


And that's where you get the impact on society wrong. The OpenAI affair shows what happens when rising up to the level of "philosophy and positive actionable visions" conflicts with the grubby, dirty, filthy lucre tackiness. The tackiness wins.




I am not sure what you are talking about. The OpenAI affair was, in terms of my compass, Altman (closer to d/acc) fighting AI safetyists from EA structures. What tackiness won? Do you mean the promise of compensations to technical staff, or the struggle over the corporate board's power? This is all instrumental to much bigger objectives.


				

				

						

							

								
							

						


						
					

				
			


			

	


			
			
				

			
		

	






		

			
			
				Culture War Roundup for the week of November 13, 2023
			
		

	

	

	

		
			

			

				

			

		

			


				

					
How's it going my dude?


				

				

						

							

								
							

						


						
					

				
			


			

	


			
			
				

			
		

	






		

			
			
				Culture War Roundup for the week of November 13, 2023
			
		

	

	

	

		
			

			

				

			

		

			


				

					
Cynicism is a cope of dysfunctional people.


				

				

						

							

								
							

						


						
					

				
			


			

	


			
			
				

			
		

	






		

			
			
				Israel-Gaza Megathread #2
			
		

	

	

	

		
			

			

				

			

		

			


				

					


It's an outpost of my civilization, organized on principles I agree with, inhabited by people I could live at ease with. They defend themselves, their borders, their interests, with the kind of vigor and thick-skinned determination I'd like to see my nation display. (If only!) I admire them and wish them well.




John Derbyshire most likely cannot live in Israel or among a representative sample of Israelis any more than he can live among immigrants to the West he hates so much.


I increasingly suspect that the only correct decision for any sane person is converting to Judaism or at the very least relinquishing any claim to being white, because Christian whites are just brain-damaged and cannot tell a universalist ideology (even "nationalism") from a population's game-theoretically advantageous modus operandi. Strong vibe of round-headed Slavic Hitlerists.


It is absurd to assume that a more competent entity sharing your material interests is your ally rather than a competitor.


				

				

						

							

								
							

						


						
					

				
			


			

	


			
			
				

			
		

	






		

			
			
				Culture War Roundup for the week of August 28, 2023
			
		

	

	

	

		
			

			

				

			

		

			


				

					
I don't really want to entangle those but might respond to a DM.


				

				

						

							

								
							

						


						
					

				
			


			

	


			
			
				

			
		

	






		

			
			
				Culture War Roundup for the week of July 17, 2023
			
		

	

	

	

		
			

			

				

			

		

			


				

					


Isn’t that the straightforward point of the gyges story?




The point of Gyges story is that, as I have said, «morality does not exist and does not differ from conventional etiquette in some substantial objective sense». It is not a singular real thing that exists outside us, it is an arbitrary system of behavioral guidelines that differ between groups and individuals. In the same post, I give my perspective on what we have learned about subjective morality and its difference from mere etiquette since Plato. Like most people, I have subjective morality that is essentially intuitive and deontological. Greeks, in Plato's cynical imagination, believed that «the best is to do injustice without paying the penalty», such was their intuitive internal moral compass, with prosociality a burdensome etiquette indeed. I live millenia later, and my idea of the highest good is different, because I am a product of biological and cultural selection pressures rewarding other priors; i have internalized a version of prosociality to a certain extent. I am not closer to some objectively established moral truth on account of this, I'm just me, doing and perpetuating what I like for contingent historical reasons. I could collapse those philosophical complexities into do's and don'ts – and I do, when speaking to children. But if being actually mush-headed and deluded about this fact is a criterion for admission into the Goodbot Society, instead of being attacked as a defector… Well, then I can only despise goodbots harder, see them as even more of immature cattle with heritable neural crest defects, and, against my better judgement, wish them more abuse at the hands of actual defectors – who delight in mouthing mawkish falsehoods, and who reshape your moral code as they see fit through revolting identity-coded morality plays that you myopically mistake for entertainment and education.


We can pick a yardstick like game-theoretical performance (assuming utility can be compared between agents) and compare moral codes with that as our source of objectivity, but this would depend on evaluation conditions; people can argue persuasively that even antisocial punishment is rational, and thus good. The issue with selecting game theory itself as one's moral truth is that you'd need widespread near-omniscience to get out of some horrible local minima, while defection-heavy heuristics remain robust (as do cooperation-heavy ones, while conditions are not horrible). Maybe there exists a perfect solution that maximizes winning across all environments even for avg 100 IQ humans dealing with incomplete information about reward scheme and unknown number of sophisticated defectbots in the environment. I do not know it, but even if it did exist, it'd only be objectively correct given our arbitrarily selected (but very handy) benchmark; and for me asesthetics matter about as much. Truth, of course, is also a matter of aesthetics.


This should clarify my position enough.




Although I’m on the record condoning the murder of any and all authority figures in nazi germany, so if I was who you think I am, we don’t even disagree on that




Good. Hopefully you realize that I take seriously what you say about your methodology for evaluating «signals», not as an edgy maximalist posture (even though you throw that in too). I know people who blithely speak and live by your code where success at attaining power justifies itself, and consider them being alive and unbothered my personal failing, as well as collective failure of my civilization.




I think you could benefit from adopting a less subjective view of morality




Well, here's what I think: you can admit that you are indeed no better than Nazi authority figures, or admit your «signals» methodology is laughably bad and does not result in nice conclusions like conventional Western morality as you want it to so should be ditched; or, indeed, just stop.




so I’ll make this the last time we talk




Good luck.


				

				

						

							

								
							

						


						
					

				
			


			

	


			
			
				

			
		

	






		

			
			
				Culture War Roundup for the week of July 10, 2023
			
		

	

	

	

		
			

			

				

			

		

			


				

					
https://youtube.com/watch?v=65zOlQV1qto&t=1854


Looking back at it I've read too much into his words. I do think it's a possible interpretation though.


				

				

						

							

								
							

						


						
					

				
			


			

	


			
			
				

			
		

	






		

			
			
				Small-Scale Question Sunday for July 9, 2023
			
		

	

	

	

		
			

			

				

			

		

			


				

					
@Amadan is it fair to call this gibberish? I suppose he can't be thinking he makes a solid point.


				

				

						

							

								
							

						


						
					

				
			


			

	


			
			
				

			
		

	






		

			
			
				Culture War Roundup for the week of July 3, 2023
			
		

	

	

	

		
			

			

				

			

		

			


				

					
I think behavioral zombies defined as such are just not interesting in the age of LLMs. It doesn't take much to fool people.


A subtler hypothetical subtype of a behavioral zombie that actually precisely matches a specific person's behavior – that is not pre-recorded but generated by zombie's own causality in the same situations – might be interesting though, and I think amounts to the neurological one, or contains it somehow.


				

				

						

							

								
							

						


						
					

				
			


			

	


			
			
				

			
		

	






		

			
			
				Small-Scale Question Sunday for July 9, 2023
			
		

	

	

	

		
			

			

				

			

		

			


				

					
Grasping at straws.


				

				

						

							

								
							

						


						
					

				
			


			

	


			
			
				

			
		

	






		

			
			
				Culture War Roundup for the week of July 3, 2023
			
		

	

	

	

		
			

			

				

			

		

			


				

					


In this case, the LLM could perfectly well be a P-zombie




No.


An LLM does not, as far as we know, employ an actual physical human brain for computation. A [strong version of] p-zombie does, its causal chains are exactly the same as in our brans, it not an arbitrary Turing-test-passing AI. I think that it "feels like something" to be an LLM computation too, but it very likely doesn't feel like having human quale.


It is obviously unwarranted to say that a system that can ape a human with its behaviors computes a human mind or any part thereof, humans can have low standards among other reasons. And in general, our external behaviors are a low-dimensional lossy and noisy projection of our internal states, so the latter cannot be fully inferred from the former, at least in realistic time (I think).


My argument hinges on the fact that a brain contains events that, from an information perspective, suffice to be described as quale with regard to other events (that are described as sensations). It is coherent to speculate that e.g. there is such a thing as an immaterial human soul and that it does not parse these events, and instead works in some other way. It is not coherent to say that they exist but also don't exist.


				

				

						

							

								
							

						


						
					

				
			


			

	


			
			
				

			
		

	






		

			
			
				Culture War Roundup for the week of June 5, 2023
			
		

	

	

	

		
			

			

				

			

		

			


				

					
Thank you for providing an example. Yes, the point of voting Trump is to Own The Libs, drive them mad. This is exactly what Hanania is talking about.




Like Mitt "the Mormon Theocrat" Romney? Being nice and clean-cut never stopped the attacks from the Democrats




I suppose we will never know how well Romney would've handled those attacks were he to become POTUS.


				

				

						

							

								
							

						


						
					

				
			


			

	


			
			
				

			
		

	






		

			
			
				Culture War Roundup for the week of May 8, 2023
			
		

	

	

	

		
			

			

				

			

		

			


				

					
Condescension is merited if you're asking for it; I mentioned IHNMAIMS in my post and this scenario is materially different from «you might be tortured and killed by a crazy person» in that it posits biological immortality.


I happen to think that a superintelligent being will at the very least be able to greatly prolong the life of its victim, this doesn't conflict with any part of our understanding of «the matter at hand» (we're pretty sure that a healthy brain can outlive the body if you provide it a decent environment). And of course this conjecture, while unproven, is vastly less far-fetched than continuity of identity between a human and a simulacrum built from indirect evidence.


				

				

						

							

								
							

						


						
					

				
			


			

	


			
			
				

			
		

	






		

			
			
				Culture War Roundup for the week of May 8, 2023
			
		

	

	

	

		
			

			

				

			

		

			


				

					
I really recommend reading the «I have no mouth and I must scream», or at least the synopsis.


				

				

						

							

								
							

						


						
					

				
			


			

	


			
			
				

			
		

	






		

			
			
				Culture War Roundup for the week of May 8, 2023
			
		

	

	

	

		
			

			

				

			

		

			


				

					
Well, it adds the important dimension of unfriendly AI being spiteful and making a friend-enemy distinction. Actually, let's just scratch the whole alignment part and say Conjecture or Anthropic announce that, should they build a God-like AI, they'll go after people who tried to get in their way and torture them, while rewarding allies with extra share of the light cone. If they were very close to success, this would've been straightforwardly persuasive.


				

				

						

							

								
							

						


						
					

				
			


			

	


			
			
				

			
		

	






		

			
			
				Culture War Roundup for the week of May 1, 2023
			
		

	

	

	

		
			

			

				

			

		

			


				

					


humans have a moral skeleton, innate hardwiring that allows us to learn morality




I suppose we have, to some extent, but it can't be all that robust. Tons of species are psychopathic by our standard, and of course this standard exists to distinguish humans who don't fit it. So it's more like a slight inductive bias, in the same way we have biases to learn to navigate in 3D space and prefer sugary foods. Biases of the algorithm can be substituted with biases in the data.




and believe it (as opposed to mimic it). This is highly instrumentally non-convergent and probably needs to be coded into an AI directly; gradient descent on output will only produce lying psychopaths mimicking morality.




I don't see why that would be true. Indeed, I do not see why gradient descent wouldn't be much better of learning deep cognitive regularities including morality. You seem to hold that morality is something essential, some set of terminal value-influences, but why is that true for morality and not any other aspect of our cognition, both instrumentally worthwhile and «instrumentally non-convergent» ones? Every part of our decision-making feels profoundly qualitatively colored for us.


Why is "coded directly" better than learned? The major reason we're doing this stuff is that it's vastly better at generalization,


Sorry, this looks like a vibe-based argument, where neural nets deceptively "imitate" and hard code is "good and honest". It's all algorithms. Inasmuch as human minds are computable, our morality is an algorithm too.




GOFAI has some hope because we could code morality directly




What good would that do? It'd break OOD just the same, and if it didn't break, it'd be rewritten or worked around by the purported daemon of optimization.




But I'm not seeing any hope of success on non-uploads without the ability to look inside the box. This is because "is moral" and "is pretending to be moral successfully" have identical output except in situations where dropping the pretence is worth it i.e. situations where there's a high chance of you losing control upon betrayal.




Reminder that LLMs cannot learn to do the «treacherous turn» because the dreaded SGD mercilessly ablates cognitive routines that do not contribute to decreasing loss in training. This, of course, holds in the general case.


But even beyond that, outputs may be similar but activations aren't, we know how to look at activations, and we know there are differences between the model subjectively evaluating its output as true or false.




it seems like the difficulty of determining whether spaghetti-code does X is generally at least as high as the difficulty of writing code that does X, which implies that making safe NNs is at least as hard as writing GOFAI




No, generation is always vastly simpler than classification unless you require classification that reconstructs the process of generation, of course.


I've long held that Yuddist program is, in addition to all else, an attractor  for a particular variation of anxiety/OCD disorder: fetishization of «being in control», of making thoughts play by «proper rules».  But it's fetishization because it doesn't really work, it pursues ghosts, precisely the deceptive external form of reliability. You gain clarity of ideas by testing them against evidence, not by being real suspicious of disconfirmations.


Pitts wrote that his depression might be “common to all people with an excessively logical education who work in applied mathematics: It is a kind of pessimism resulting from an inability to believe in what people call the Principle of Induction, or the principle of the Uniformity of Nature. Since one cannot prove, or even render probable a priori, that the sun should rise tomorrow, we cannot really believe it shall.”


This malady is to be treated, not nurtured.