--- Log opened Thu Jun 02 00:00:40 2022
01:28 -!- darsie [~darsie@84-113-55-200.cable.dynamic.surfer.at] has joined #hplusroadmap
01:47 -!- lkcl [~lkcl@92.40.174.171.threembb.co.uk] has quit [Quit: Leaving]
03:01 -!- Molly_Lucy [~Molly_Luc@user/Molly-Lucy/x-8688804] has quit [Read error: Connection reset by peer]
03:06 -!- Molly_Lucy [~Molly_Luc@user/Molly-Lucy/x-8688804] has joined #hplusroadmap
03:15 -!- Molly_Lucy [~Molly_Luc@user/Molly-Lucy/x-8688804] has quit [Quit: Textual IRC Client: www.textualapp.com]
03:34 -!- adlai is now known as adlai|alphanumon
03:34 -!- adlai|alphanumon is now known as adlai|alphanum
03:35 -!- Molly_Lucy [~Molly_Luc@user/Molly-Lucy/x-8688804] has joined #hplusroadmap
03:40 < kanzure> https://fantasticanachronism.com/2021/03/23/two-paths-to-the-future/
03:41 < kanzure> "I believe the best choice is cloning. More specifically, cloning John von Neumann one million times."
03:42 < kanzure> "There are commercial dog- and horse-cloning operations today. We've even cloned primates. The costs, frankly, are completely trivial. A cloned horse costs $85k! Undoubtedly the first JvN would cost much more than that, but since we're making a million of them we can expect economies of scale. I bet we could do it for $200 billion, or less than half what the US has spent on the F35 thus far. ...
03:42 < kanzure> ...Compared to the predicted cost of training superhuman AGI, the cost of one million JvNs is at the extreme lower end of the AI forecasts, about 5 orders of magnitude above GPT-3."
03:43 < kanzure> "Assuming interests are highly heritable we should probably also re-create leading biologists, chemists, engineers, entrepreneurs, and so on. No bioethicists though."
03:59  * nsh strikes the Poe's Law gong
04:01 < nsh> .wik The Alignment Problem
04:01 < saxo> "The Alignment Problem: Machine Learning and Human Values is a 2020 non-fiction book by the American writer Brian Christian. It is based on numerous interviews with experts trying to build artificial intelligence systems, particular machine learning systems, that are aligned [...]" - https://en.wikipedia.org/wiki/The_Alignment_Problem
04:02 -!- Molly_Lucy [~Molly_Luc@user/Molly-Lucy/x-8688804] has quit [Read error: Connection reset by peer]
04:09 < kanzure> here is a short story about artificial wombs from someone (seems a little boring to me, but i don't think it's trying to be non-boring?) https://docs.google.com/document/d/10qaBFOWYVJPVWMq9rMJ3wFX3hDns7-_XHnlouj1S5so/edit
04:13 < kanzure> https://atis.substack.com/p/the-world-in-2072 "So what caused everyone to grow slightly richer? The answer is the boring one. Things gradually got better over time through new technologies (not one single earth-shattering technology) and innovations, and more people are well-educated enough to be very productive in the world economy. That’s it, the secret sauce is just education and ...
04:13 < kanzure> ...compounding returns over time. Again, many opinion pieces were written."
04:17 -!- Molly_Lucy [~Molly_Luc@user/Molly-Lucy/x-8688804] has joined #hplusroadmap
04:18 < kanzure> https://ftxfuturefund.org/principles/ "Humanity’s future could be vast. Our species could survive for hundreds of millions of years, until the Earth is no longer habitable, or for even longer. Countless people might lead lives of flourishing or misery—or never live at all—depending on what humanity does today. We want to fund projects with clear answers to the question: “How will this ...
04:18 < kanzure> ...project improve humanity’s odds of surviving and flourishing for thousands of years or longer?”"
04:28 < nsh> if you boof enough of the longnowthink pipe you eventually go back to living in the moment and the species prospers by default
04:29 < nsh> trying to outsmart the universe that created you is an interesting journey but the destination is the same as the origin
04:29 < nsh> with respect to humanity extinction is a conceptual negative space
04:30 < nsh> the universe abhors the vacuum and it would take cosmic energies to prevent it humaning
04:30 < nsh> survivalists are just hobbyists whose horse is recreational worrying
04:31 < nsh> which is itself a feature of the aforementioned state of affairs
04:32 < nsh> the proposition most worthy of examination is that of whether the outcome being in doubt at all is actually essential to the process proceeding
05:40 < muurkha> most species are already extinct; survival is not the default
05:41 < muurkha> nmz787: "cheaper than gravel" sort of implies that you'd have to manufacture it in situ, because anything you hauled in from elsewhere would be more expensive than gravel
05:42 < muurkha> well, as expensive as gravel
05:49 -!- oxphi [~oxphi@107.181.189.44] has joined #hplusroadmap
06:06 -!- yashgaroth [~ffffffff@2601:5c4:c780:6aa0::93] has joined #hplusroadmap
06:11 < nsh> if a runner leaves a slipper behind, is there still a forward motion?
06:12 < nsh> or when a snake sheds its skin is there still a snake?
06:19 < kanzure> really unfortunate that all of this energy is focused on x-risk reduction and altruism instead of building and trade https://theprecipice.com/ https://www.cold-takes.com/most-important-century/ 
06:21 < muurkha> I think there's already a lot more energy focused on building and trade than on x-risk reduction and altruism.  off the cuff I'd guess about 2.5 orders of magnitude more
06:22 < muurkha> for building.  4 orders of magnitude more for trade
06:22 < kanzure> x-risk reduction isn't a productive pursuit because anyone who can successfully reduce x-risk can also successfully increase x-risk
06:23 < muurkha> your conclusion does not follow from its premise; its opposite does
06:23 < kanzure> there needs to be a positive sentiment of the future that is not rooted in altruism
06:23 < muurkha> loosely, at least
06:24 < kanzure> the "(somehow-)inarguably omnipotent" x-risk class is not worth our time arguing
06:24 < muurkha> trade is currently US$85 trillion per year
06:25 < kanzure> i assure you that most of that trade is not focused on million year deep human flourishing
06:25 < muurkha> assuredly true
06:26 < muurkha> but million-year-deep human flourishing is an altruistic aim unless you manage to live a million years yourself
06:26 < muurkha> and even then it's mostly altruistic
06:26 < kanzure> most people do not consider exerting influence over million year timescales to be altruistic
06:26 < kanzure> and hence why there are common law limitations on the lifespan of "indefinite" trusts thru inheritance
06:26 < muurkha> most people don't consider anything at all
06:27 < muurkha> altruism doesn't give you a blanket exemption from the common law, so your reasoning is invalid
06:30 < muurkha> the fact is that people exerting influence inevitably come into conflict with other people trying to exert influence, and when those people have been dead for generations, they are in a worse position to defend themselves, so they tend to lose political battles
06:30  * nsh smiles
06:32 < nsh> though oft the ghost a sharp riposte reveals
06:32 -!- lkcl- is now known as lkcl
06:35 < lkcl> saxo: the problem with AI is that the understanding of neurons is fundamentally f****d, and has been for 40+ years.
06:36 < muurkha> lkcl: heh, nice joke
06:36 < lkcl> it's best explained by my friend wai tsang, who has made a life-long study of the brain, with a view to implementing it. https://www.youtube.com/watch?v=Q7ThBK33Svk
06:36 < kanzure> there is some irony in arguing with a bot on the topic of artificial intelligence
06:36 < muurkha> #thatsthejoke
06:36 < kanzure> .title
06:36 < saxo> A new, very consequential & huge scientific idea. Like E=MC2  Life = Intelligence to True AI - YouTube
06:36 < lkcl> muurkha, hah, i forget that saxo's the bot in this channel :)
06:37 < nsh> .t
06:37 < saxo> A new, very consequential & huge scientific idea. Like E=MC2  Life = Intelligence to True AI - YouTube
06:37 < lkcl> because it's not named "bot"
06:37 < nsh> oopss soz
06:37 < lkcl> nsh, kanzure, thx :)
06:37 < kanzure> there needs to be more work on human brain simulation instead of emulation
06:38 < lkcl> https://youtu.be/Q7ThBK33Svk?t=913 for example
06:38 < kanzure> there are only so many high-level pathways of various cortex layers interconnecting, and they are somewhat common between mammalian brain architectures
06:38 < lkcl> kanzure, the problem is that the study of the neurons themselves is not being properly forwarded to people-working-on-simulation/NNs
06:39 < lkcl> for example, it's not even understood in the standard AI community that neurons can perform boolean logic *including XOR*
06:39 -!- mrdata [~mrdata@user/mrdata] has quit [Ping timeout: 258 seconds]
06:39 < kanzure> there's a lot of good neurophysiology out there, sure
06:39 < lkcl> and it's *definitely* not understood that neurons can perform both differentiation and integration with respect to time
06:40 < lkcl> the entire fundamental assumption of AI NNs is: "take the sum of the inputs and do s*** with it"
06:40 < muurkha> really?  I thought basic physical considerations would make it obvious that neurons integrate over time
06:40 < kanzure> it's true that computational models of neurons have further to develop, although i'm not sure we need well-modeled neurons to get interesting behaviors out of these systems
06:40 < lkcl> which is about as deeply flawed as it can possibly get
06:41 < lkcl> the other assumption is that the dendrons only extend to a small localised area (the bits that "fire other neurons")
06:41 < muurkha> I don't think anyone working on ANNs believes their "neurons" are even vaguely accurate models of actual neurons
06:41 < lkcl> reality is that the dendrons of most neurons extend vast distances, distributed throughout all other cortexes
06:42 < lkcl> muurkha, and yet, they get billions of dollars of investment to create something that cannot say "no that's neither an apple nor an orange"
06:42 < lkcl> they attempt to classify it as a "borange"
06:42 < lkcl> (mathematically, i mean)
06:43 < lkcl> also the whole idea of "training" using 100,000 samples is completely f****g ridiculous
06:43 < muurkha> pretty sure image recogition ANNs have been able to distinguish between more than two classes of things for at least 50 years
06:43 < kanzure> "How thalamic relays might orchestrate supervised deep training and symbolic computation in the brain" https://www.biorxiv.org/content/10.1101/304980v1.abstract
06:44 < lkcl> anyone who's studied brain physiology knows that neural structures only require *one* example (one experience) to "learn"
06:44 < lkcl> all other "experiences" become comparative differential *additional* experiences that refine that initial first learning experience.
06:44 < lsneff> We’re so far away from brain simulation
06:44 < muurkha> they've been getting billions of dollars of investment because they've built systems that work in practice, not because they think they're close to brain simulation
06:45 < lsneff> Neuron models are wayyy too primitive
06:45 < lkcl> muurkha, the problem isn't what the network is trained *to* recognise, the problem comes when you show such a deep-trained network something it *hasn't* been shown before
06:45 < lkcl> train on apple
06:45 < lkcl> train on orage
06:45 < lkcl> no problem
06:45 < lkcl> show it a mouse and it literally goes "that's a borange"
06:45 < lsneff> And there are plenty of things in the brain other than neurons that play an active role
06:45 < muurkha> no, neural networks do not 'literally go "that's a borange"'
06:46 < lkcl> lsneff, indeed. mesolimbic dopamine system, chemicals that affect the supply, etc. etc.
06:46  * nsh notes that science is well past any one given scientist actually comprehending what they're doing at the frontier
06:46 < muurkha> lsneff: what do you think of OpenWorm?
06:46 < lkcl> muurkha, i'm simplifying :)
06:46 < kanzure> or nemaload if not openworm
06:46 < muurkha> lkcl: the problem is that your simplification is wrong, which is indistinguishable from you not knowing what you're talking about :)
06:47 < lkcl> muurkha, ngggh... /me hand-waving.... true... ish :)
06:47 < lkcl> i have resisted going into depth on it because i know i won't come back out of a study of the topic, for a decade
06:48 < lkcl> consequently i know enough to know there's something "wrong" but not enough to explain it fully to someone who *does* know the current research, in-depth, if that makes sense
06:49 < lkcl> hence why i'm going off of what my friend wai has done (paraphrasing, inaccurately) because he *has* spent the past... err... 35 years studying neuroscience
06:50 < muurkha> there are accessible overviews of current ANNs and quite a bit of open-source tooling; it would probably be worthwhile for you to spend, if not a decade, at least a few hours studying it
06:51 < muurkha> current ANNs do not attempt to be accurate models of neuroscience but they are the best available solutions for a wide variety of heuristic problems, and the optimization techniques they use are even more broadly applicable than the ANNs themselves
06:52 -!- ^ditto [~limnoria@crap.redlegion.org] has quit [Remote host closed the connection]
06:52 < muurkha> you might be able to apply them to problems that are of interest to you after a few dozen hours of investigation
06:52 < lkcl> muurkha, my interest lies in actually tackling the area of proper and full brain / consciousness implementation
06:53 < lkcl> for that purpose, anything that attempts to only do "heuristics" is of no value, other than as a lesson of what *not* to do
06:54 < lkcl> as in: i don't have any interest, at all, in any of the "standard" uses to which ANNs are applicable
06:54 < kanzure> here is my library:
06:54 < kanzure> https://diyhpl.us/~bryan/papers2/ai/machine-learning/
06:54 < kanzure> i haven't been including recent works tho, so recommendations appreciated
06:55 < lkcl> the one that got me really excited recently was an ASIC by Intel that is capable of learning from a single example/experience
06:56 < lkcl> *that's* the sort of thing i'm interested to find out how they did it
06:56 < lkcl> because it's half-way to proper self-awareness / consciousness
06:58 < muurkha> yeah, one-shot learning is pretty exciting
06:59 < muurkha> https://proceedings.neurips.cc/paper/2016/file/90e1357833654983612fb05e3ec9148c-Paper.pdf is a NeurIPS (NIPS?) paper on the subject from 02016 which Google Scholar tells me has 4294 citations
07:00 < muurkha> I am amused to note that of the five authors four of them have Google email addresses that are based on their (legal?) names, but the fifth is countzero@google.com
07:01 < lkcl> https://www.intel.co.uk/content/www/uk/en/research/neuromorphic-computing.html
07:01 < lkcl> intel neuromorphic asics, codenamed "loihi"
07:01 < muurkha> okay but that's PR, I linked you to actual research
07:02 -!- mrdata [~mrdata@user/mrdata] has joined #hplusroadmap
07:02 < lkcl> i'm a few steps behind! :)
07:03 < muurkha> kanzure's library is well worth perusing
07:04 < muurkha> also that page seems to be about SNNs rather than one-shot learning?
07:07 -!- spaceangel [~spaceange@ip-78-102-216-202.net.upcbroadband.cz] has joined #hplusroadmap
07:18 < lsneff> Loihi is a pretty modular SNN array
07:18 < lsneff> I read a paper about implementing astrocyte analogs in loihi
07:22 -!- Molly_Lucy [~Molly_Luc@user/Molly-Lucy/x-8688804] has quit [Read error: Connection reset by peer]
07:29 < muurkha> neat
08:12 < oxphi> muurkha: which library?
08:17 < muurkha> 13:54 < kanzure> here is my library:
08:17 < muurkha> 13:54 < kanzure> https://diyhpl.us/~bryan/papers2/ai/machine-learning/
08:23 -!- mrdata [~mrdata@user/mrdata] has quit [Ping timeout: 256 seconds]
08:40 < oxphi> thx
09:12 < maaku> lkcl: modern AI is not trying to be bio-accurate
09:12 < maaku> and there's no reason to suppose it ought to be
09:22 -!- Molly_Lucy [~Molly_Luc@user/Molly-Lucy/x-8688804] has joined #hplusroadmap
09:36 -!- HD36079 [~joey@m90-131-37-205.cust.tele2.lt] has joined #hplusroadmap
09:36 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has left #hplusroadmap []
09:50 -!- HD36079 [~joey@m90-131-37-205.cust.tele2.lt] has quit [Quit: Leaving]
09:54 -!- HD36079 [~joeyskmat@2001:470:69fc:105::2:2143] has joined #hplusroadmap
09:56 -!- Molly_Lucy [~Molly_Luc@user/Molly-Lucy/x-8688804] has quit [Read error: Connection reset by peer]
10:30 -!- Molly_Lucy [~Molly_Luc@user/Molly-Lucy/x-8688804] has joined #hplusroadmap
10:44 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has joined #hplusroadmap
11:19 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has quit [Ping timeout: 276 seconds]
11:43 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has joined #hplusroadmap
11:49 -!- mrdata [~mrdata@user/mrdata] has joined #hplusroadmap
12:15 -!- ^ditto [~limnoria@crap.redlegion.org] has joined #hplusroadmap
12:15 -!- ^ditto [~limnoria@crap.redlegion.org] has quit [Client Quit]
12:16 -!- ^ditto [~limnoria@crap.redlegion.org] has joined #hplusroadmap
14:09 < fenn> most x-risks are not inarguably omnipotent: plague, asteroids, nuclear war, climate change
14:33 < kanzure> mutually assured destruction is surely a kind of omnipotence that defeats everyone right?
14:33 < kanzure> i suppose we never got evidence that countries would be actually effective at implementing true mutually assured irradiation of the planet
14:46 < maaku> Gee, I wish we had one of them doomsday machines.
14:52 < kanzure> how about an anti-doomsday machine instead?
14:57 < maaku> sorry was a quote from Dr. Strangelove
14:58 < maaku> I want a Star Wars global anti-ICBM shield
15:03 < L29Ah> i want my aliexpress orders delivered with mass drivers
15:51 -!- spaceangel [~spaceange@ip-78-102-216-202.net.upcbroadband.cz] has quit [Remote host closed the connection]
16:01 < kanzure> backup your data
16:02 < L29Ah> i do it daily automatically, and you should
16:22 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has left #hplusroadmap []
16:35 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has joined #hplusroadmap
18:07 -!- darsie [~darsie@84-113-55-200.cable.dynamic.surfer.at] has quit [Ping timeout: 258 seconds]
18:16 -!- Llamamoe [~Llamamoe@178235178120.dynamic-4-waw-k-1-2-0.vectranet.pl] has quit [Quit: Leaving.]
18:56 -!- yashgaroth [~ffffffff@2601:5c4:c780:6aa0::93] has quit [Quit: Leaving]
19:39 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has quit [Ping timeout: 255 seconds]
20:03 -!- bulbasaur [~bulbasaur@207.253.236.155] has quit [Remote host closed the connection]
20:06 -!- bulbasaur [~bulbasaur@207.253.236.155] has joined #hplusroadmap
20:15 < maaku> test your backups
20:47 -!- mlaga97 [~quassel@user/mlaga97] has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
20:47 -!- mlaga97 [~quassel@user/mlaga97] has joined #hplusroadmap
--- Log closed Fri Jun 03 00:00:41 2022