--- Log opened Sat Nov 02 00:00:38 2024
00:33 -!- TMM [hp@amanda.tmm.cx] has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
00:33 -!- TMM [hp@amanda.tmm.cx] has joined #hplusroadmap
00:59 < hprmbridge> kanzure> https://www.lesswrong.com/posts/jTiSWHKAtnyA723LE/overview-of-strong-human-intelligence-amplification-methods
01:51 -!- darsie [~darsie@84-112-12-36.cable.dynamic.surfer.at] has joined #hplusroadmap
04:27 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has joined #hplusroadmap
05:05 -!- millefy [~Millefeui@91-160-78-132.subs.proxad.net] has quit [Read error: Connection reset by peer]
05:07 -!- millefy [~Millefeui@91-160-78-132.subs.proxad.net] has joined #hplusroadmap
05:09 -!- millefy [~Millefeui@91-160-78-132.subs.proxad.net] has quit [Read error: Connection reset by peer]
05:10 -!- millefy [~Millefeui@91-160-78-132.subs.proxad.net] has joined #hplusroadmap
07:35 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has left #hplusroadmap []
08:35 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has joined #hplusroadmap
09:05 < hprmbridge> kanzure> https://x.com/bayeslord/status/1852735838859186263
09:21 -!- Gooberpatrol66 [~Gooberpat@user/gooberpatrol66] has joined #hplusroadmap
09:32 < hprmbridge> alonzoc> I'm fairly suspect of any non-drastic intervention having a meaningful effect. editing or selection of embryos to maximise some heuristic expected intelligence metric feels most promising but fairly useless in adults.
09:32 < hprmbridge> alonzoc> Algernon's law seems strong to me the brain itself is fairly optimised, even if we can afford an increased metabolic cost in terms of calorie intake any large changes would requires changes to heat dissipation, blood vessel density, and the cleanup of transmitter buildup during wakefulness. Probably the composition of the brain itself given that axon diameters and the like are optimised for the
09:32 < hprmbridge> alonzoc> existing "demand" for bandwidth given energy constraints
09:35 < kanzure> embryo engineering can do better than selection
09:35 < kanzure> especially if you are not beholden to the cult of polygenic scoring
09:36 < kanzure> i don't know if anyone hears themselves when they say "but useless in adults".... like imagine we had a +20 year longevity intervention.
09:37 < kanzure> and we're like "oh well, it doesn't work on me, so fuck everyone for the next 200 years"
09:37 < kanzure> ????
09:37 < kanzure> it's not just you. it's the practically universal reaction.
09:37 < L29Ah> faster, denser and more reliable neurointerfaces are a key to dramatic intelligence enhancement
09:39 < hprmbridge> alonzoc> Oh yeah I'm pro engineering for the future my point is a lot of people are looking at this with a view for themselves. Interest in engineering their descendents is less widespread. It's why exocorticies are fairly popular they can be done to "me" "now".
09:39 < hprmbridge> alonzoc> I suspect we'll see embryo selection before full engineering and a single generation or so of embryo selection could have quite an impact
09:40 < hprmbridge> alonzoc> kanzure: What did you mean about the "cult of polygenic scoring"
09:43 < L29Ah> mass human manufacturing will shift focus towards the embryo engineering a lot, compared to the current overly-distributed(-away) means of reproduction that is logistically and organizationally annoying
09:44 < kanzure> alonzoc: awareness of polygenic scoring and polygenic traits has erased people's knowledge of genetic engineering and replaced it with concerns about >1,000 polygenic effects
09:46 < hprmbridge> alonzoc> Ahh, yeah the whole "the genetic code is beyond mortal ken" school of thought...
09:47 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has quit [Ping timeout: 260 seconds]
09:49 < hprmbridge> alonzoc> Yeah it's a headache but if you have millions of whole sequences combined with a ton of metrics you could do causal inference and produce a ensemble of candidate causal models. Then you "just" (lab work is expensive but all of this is *doable*) need to perform experiments to narrow down the causal models and well done you've now got a useful way to select for specific polygenic effects and
09:49 < hprmbridge> alonzoc> understand the consequences of your edits.
09:49 < kanzure> okay no
09:50 < kanzure> i mean you can choose to work in the polygenic regime. but why would you want to do that? it is strictly worse than the alternatives!
09:51 < kanzure> height is an excellent example. it's a polygenic trait. but it turns out that upregulating hormones- such as through straight hormone injection into children- can increase growth! it turns out that genetic expression circuits continue to be a real thing even if people are obsessed with polygenic traits.
09:52 -!- redlegion [redlegion@omghax.redlegion.org] has quit [Remote host closed the connection]
09:53 < hprmbridge> alonzoc> Well if you want to do extensive edits you are gonna have to graduate from working with single genes to modifying the whole system. If you want to change morphology of the brain in a useful way you'll need to work in a polygenic regime
09:54 < kanzure> we already have evidence of reasonably intelligent humanoids without having to substantially change human brain neuroanatomy
09:55 < kanzure> it really does look like a handful of minor tweaks makes substantial progress on human intelligence
09:56 < kanzure> although, i'm saying that from the perspective of a human, in the grand scheme of intelligent systems perhaps not so much :)
09:56 -!- redlegion [redlegion@omghax.redlegion.org] has joined #hplusroadmap
09:57 < hprmbridge> alonzoc> True but if you wanna break through the "160-200 IQ" (not concrete etc) regime you'll probably need a more extensive architectural overhaul.
09:57 < hprmbridge> alonzoc> I totally agree on the handful of minor tweaks point but it would be moving up of the mean but a large drop in standard deviation.
09:57 < hprmbridge> alonzoc> The social impact of that would be massive though
09:58 < kanzure> not so interested in social impacts on that, just individual impacts. but i guess there is some sort of eusocial benefit or something.
09:58  * kanzure goes back to https://www.youtube.com/watch?v=S-Jo-djilvo
10:02 < hprmbridge> alonzoc> Wow I'll have to add this to my collection alongside the ballad of big yud
10:03 -!- TMM [hp@amanda.tmm.cx] has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
10:03 -!- TMM [hp@amanda.tmm.cx] has joined #hplusroadmap
10:03 < kanzure> there is also the (more silly) elon one https://www.youtube.com/watch?v=jvSJORwb134
10:05 < hprmbridge> alonzoc> I suspect there would be benefit to education that'd probably allow more people to reach their phenotypic maximum
10:06 < hprmbridge> alonzoc> On the topic of single edits wasn't their some chatter that the chinese HIV experiment was partially motivated because the gene increased intelligence in rats
10:07 < kanzure> there was some yapping about that as a possible effect of the edit, i forget the conclusion on whether that edit was actually cognitorelevant
10:08 -!- Gooberpatrol66 [~Gooberpat@user/gooberpatrol66] has quit [Quit: Konversation terminated!]
10:10 -!- Gooberpatrol66 [~Gooberpat@user/gooberpatrol66] has joined #hplusroadmap
10:17 < hprmbridge> alonzoc> Is the human-gigafactory page a new one?
10:17 < kanzure> yes it's from a recent tweet and a helpful wikihelper put it on the wiki
10:18 < kanzure> https://gnusha.org/logs/2024-10-28.log
10:18 < hprmbridge> alonzoc> Ahh an IQ printer to feed into the IQ shredders 😄 /hj
10:25 < hprmbridge> alonzoc> > so here's where my understanding of the statistics falls apart
10:25 < hprmbridge> alonzoc> > since IQ is highly heritable, we're shifting the high end of the distribution. therefore we should expect more geniuses per selected embryo than merely adding to the average
10:25 < hprmbridge> alonzoc> > but how do you actually plug numbers into this?
10:25 < hprmbridge> alonzoc>
10:25 < hprmbridge> alonzoc> Answer to this for fenn the simplest low-order aproximation for this can be taken from "fisher's fundamental theorem of natural selection".
10:25 < hprmbridge> alonzoc> The rate of change in a populations mean fitness is proportional to it's variance.
10:25 < hprmbridge> alonzoc> You do get a compounding effect over time but if you are purely myopic optimising for your objective locally per embryo you reach a point of diminishing returns if you only rely on stronger selection pressures via embryo selection.
10:25 < hprmbridge> alonzoc> You need an explicit variance/exploration term if you proceed with just embryo selection
10:50 -!- Gooberpatrol66 [~Gooberpat@user/gooberpatrol66] has quit [Ping timeout: 252 seconds]
11:22 < docl> Instead of limiting outselves to once per generation, it would be nice to do something that makes it possible to alter one's genome later in life. This adds safety and thus permits riskier approaches, as you can roll back any bad changes.
11:23 < kanzure> genomic landing pads are on the list.
11:42 -!- RangerMauve [m-4bpbmo@matrix.mauve.moe] has quit [Remote host closed the connection]
11:44 -!- RangerMauve [m-4bpbmo@matrix.mauve.moe] has joined #hplusroadmap
11:47 < docl> Not just about risk though, there's also opportunity cost. You can only test so many novel genes per person if it's write-once, but if it's rewritable, you can test more possibilities over time per individual (and can also do more of the testing in adults, in small numbers of cells in vivo).
11:49 -!- justanotheruser [~justanoth@gateway/tor-sasl/justanotheruser] has quit [Remote host closed the connection]
11:50 -!- justanotheruser [~justanoth@gateway/tor-sasl/justanotheruser] has joined #hplusroadmap
11:57 < kanzure> https://www.telegraph.co.uk/films/2024/11/02/designer-babies-joy-netflix-film/
12:01 -!- Gooberpatrol66 [~Gooberpat@user/gooberpatrol66] has joined #hplusroadmap
13:08 -!- Gooberpatrol66 [~Gooberpat@user/gooberpatrol66] has quit [Quit: Konversation terminated!]
14:30 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has joined #hplusroadmap
15:57 < fenn> we can just assume the new population is also a normal distribution with a shifted average, and then calculate the new distance in standard deviations from the new mean (to "genius", whatever that is)
15:58 < fenn> but you have to use the old standard deviations from the general population because that's how IQ is defined
16:00 < fenn> docl it doesn't work that way. there is a developmental process orchestrated by the genome, and once that's done, it's done. you can then do remodeling later on i guess, but it won't have the same effects as starting with a different body plan in the first place
16:00 < hprmbridge> alonzoc> fenn: You can assume the distribution stays normal for a simple model. The important part is that selection increases the mean on some metric but also reduces population variance depending on how strong your selection mechanism is, you have a exploit/explore tradeoff. You can keep IQ defined with respect to the current population as a baseline.
16:02 < fenn> uh, right. the new population actually has a different standard deviation
16:05 < hprmbridge> alonzoc> Yeah which reduces the rate of change of the mean of the population which gives you a diminishing returns effect. A proper model needs to consider how genome space maps to the IQ metric but if you want a simple model you can just play around with a fixed normal distribution and the rate of change of the mean proportional to variance which decreases as a function of your selection strength.
16:10 -!- etc-vi [~etc-vi@user/meow/girlchunks] has quit [Quit: nya]
16:11 -!- etc-vi [~etc-vi@user/meow/girlchunks] has joined #hplusroadmap
16:20 < hprmbridge> alonzoc> On a tangent, gaussians have been bothering me for a while. I've seen two seperate apparently unrelated reasons gaussians pop up everywhere. The first is that they are the highest entropy distribution you can have with a fixed variance, so if you want a distribution over reals with variance `1` the maximum entropy distribution is a normal distribution with some mean. Which is fairly intuitive for
16:20 < hprmbridge> alonzoc> why they pop up, gaussians express the least possible knowledge about a variable which has finite variance.
16:20 < hprmbridge> alonzoc> On the other hand, there is the explanation from statistical physics. Given a smooth non-degenerate energy landscape all minima locally look like a quadratic potential and thus the gibbs distribution locally looks like a gaussian and approaches a gaussian globally as temperature tends to zero. (and if the landscape is degenerate so long as the ground states form a non-singular variety the
16:20 < hprmbridge> alonzoc> distribution looks like a guassian on sub-manifolds with non-degenerate hamiltonian function hessians)
16:20 < hprmbridge> alonzoc>
16:20 < hprmbridge> alonzoc> The thing that puzzles me is how these two explanations link together
16:20 < L29Ah> 09:58 < kanzure> not so interested in social impacts on that, just individual impacts. but i guess there is some sort of eusocial benefit or something.
16:20 < L29Ah> do we even have means to put high IQ people to good use? Perelman doesn't seem optimally utilized, for example
16:28 < hprmbridge> alonzoc> I'd argue Perelman's contributions aren't a waste of his talent. On the other hand promising engineers, mathematicians, and physicists who end up going into HFT or promising non-STEM people who end up in advertising i'd argue are a waste. Advertising is usually a zero-sum game or exploitative of the audience, while draining both promising engineers and humanities talent. While HFT has some
16:28 < hprmbridge> alonzoc> positive benefits but the whole game could be rendered moot by improved market design
16:51 -!- TECHNOTRON [~Malvolio@idlerpg/player/Malvolio] has quit [Quit: 2916 NEQVEPORROQVISQVAMESTQVIDOLOREMIPSVMQVIADOLORSITAMETCONSECTETVRADIPISCIVELIT 2024-11-02 23:51:05:830]
17:10 < hprmbridge> kanzure> we waste enough normal people's time with wrongthink, it's not an IQ problem
18:57 -!- darsie [~darsie@84-112-12-36.cable.dynamic.surfer.at] has quit [Ping timeout: 252 seconds]
20:58 < hprmbridge> thatchemist>  https://cdn.discordapp.com/attachments/1064664282450628710/1302481940540030976/image0.jpg?ex=67284655&is=6726f4d5&hm=373b6d5416acf75c7423910122ab215af47003df50c1e143c11f1a54351241dc&
20:58 < hprmbridge> thatchemist> ^this is my biggest finding, paired with $120 SF1000 growlights from spiderfarmer which each use ~1kWh per day
20:59 < hprmbridge> thatchemist> If you can add a catalytic converter or if exhaust is clean, then your plants accelerate growth drastically
21:00 < hprmbridge> thatchemist> Otherwise just a bucket or cup of yeast and sugar is enough to get 1200-2000 ppm CO2
21:00 < hprmbridge> thatchemist> And it increases growth at least 3-4x based on preliminary experiments
21:23 -!- mxz [~mxz@user/mxz] has quit [Ping timeout: 252 seconds]
21:24 -!- mxz [~mxz@user/mxz] has joined #hplusroadmap
21:24 -!- mxz_ [~mxz@user/mxz] has quit [Ping timeout: 244 seconds]
23:16 < hprmbridge> geraldmahony> What do you think about the proposal to put scientific publication on chain?
23:31 -!- mxz_ [~mxz@user/mxz] has joined #hplusroadmap
--- Log closed Sun Nov 03 00:00:39 2024