--- Log opened Fri Feb 09 00:00:53 2024
01:43 < hprmbridge> kanzure> "Neural abstract Reasoner" https://arxiv.org/abs/2011.09860
01:52 < hprmbridge> kanzure> https://openreview.net/forum?id=6Um8P8Fvyhl
01:53 -!- darsie [~darsie@84-113-55-200.cable.dynamic.surfer.at] has joined #hplusroadmap
03:52 -!- AugustaAva [~x@yoke.ch0wn.org] has quit [Ping timeout: 264 seconds]
03:54 < kanzure> instead of curing the FDA, what you need is more of the sickness
04:20 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has joined #hplusroadmap
06:33 -!- justanotheruser [~justanoth@gateway/tor-sasl/justanotheruser] has quit [Ping timeout: 255 seconds]
06:36 -!- justanotheruser [~justanoth@gateway/tor-sasl/justanotheruser] has joined #hplusroadmap
07:45 -!- TMM_ [hp@amanda.tmm.cx] has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
07:45 -!- TMM_ [hp@amanda.tmm.cx] has joined #hplusroadmap
08:15 -!- gptpaste [~x@yoke.ch0wn.org] has quit [Remote host closed the connection]
08:16 -!- EmmyNoether [~EmmyNoeth@yoke.ch0wn.org] has quit [Remote host closed the connection]
08:16 -!- EmmyNoether [~EmmyNoeth@yoke.ch0wn.org] has joined #hplusroadmap
08:31 -!- millefy [~Millefeui@91-160-78-132.subs.proxad.net] has joined #hplusroadmap
09:05 < kanzure> sam altman seeking $7 trillion for compute and energy for AI projects
09:05 < kanzure> millefy: welcome.
09:05 < hprmbridge> soul_syrup> typical
09:05 < hprmbridge> soul_syrup> lol
10:17 < hprmbridge> kanzure> "Blood biomarker profiles and exceptional longevity: comparison of centenarians and non-centenarians in a 35-year follow-up of the Swedish AMORIS cohort" https://link.springer.com/article/10.1007/s11357-023-00936-w
10:21 < hprmbridge> kanzure> "A genetic circuit on a single DNA molecule as an autonomous dissipative nanodevice" https://www.nature.com/articles/s41467-024-45186-2
10:35 < hprmbridge> kanzure> "Identification of pharmacological inducers of a reversible hypometabolic state for whole organ preservation" https://elifesciences.org/reviewed-preprints/93796
11:58 < hprmbridge> kanzure> https://www.asimov.press/p/synbio-guide
13:01 < kanzure> bitcoin-dev mailing list migration announcement and info: https://lists.linuxfoundation.org/pipermail/bitcoin-dev/2024-February/022327.html https://gnusha.org/pi/bitcoindev/CABaSBaxDjj6ySBx4v+rmpfrw4pE9b=JZJPzPQj_ZUiBg1HGFyA@mail.gmail.com/ https://twitter.com/kanzure/status/1756055826723217659 -> https://groups.google.com/g/bitcoindev/
14:50 < hprmbridge> kanzure>  https://cdn.discordapp.com/attachments/1064664282450628710/1205646832026394684/GF6sANqXEAAATf9.png?ex=65d9211c&is=65c6ac1c&hm=28b1bfaa18b1925dd1f5e7a93e991e616408c127aa4f51a14e06523734579d12&
14:53 < kanzure> https://teachprivacy.com/why-privacy-matters-an-interview-with-neil-richards/
15:21 -!- alethkit [23bd17ddc6@sourcehut/user/alethkit] has quit [Ping timeout: 256 seconds]
15:23 -!- alethkit [23bd17ddc6@sourcehut/user/alethkit] has joined #hplusroadmap
15:38 -!- alethkit [23bd17ddc6@sourcehut/user/alethkit] has quit [Ping timeout: 255 seconds]
15:40 -!- Goober_patrol66 [~Gooberpat@2603-8080-4540-7cfb-0000-0000-0000-113a.res6.spectrum.com] has quit [Quit: Leaving]
15:45 < fenn> i want to joke about it, but are there really people with $7 trillion to spend?
15:53 < L29Ah> dollars are cheap, hardware manufacturing capacity would be a much more pressing concern
15:55 < hprmbridge> kanzure> this is your competition https://twitter.com/rob_carlson/status/1756091625271599289
16:01 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has left #hplusroadmap []
16:14 -!- alethkit [23bd17ddc6@sourcehut/user/alethkit] has joined #hplusroadmap
16:16 < fenn> i thought the tweet was going to be about bio-engineered organoids for AI
16:16 < fenn> we could train organoids and then distill their output logits (or whatever it is) into a digital neural network
16:17 < fenn> even though there's no direct readout of the network weights
16:17 < hprmbridge> kanzure> is the argument that biological learning is faster and cheaper, but not for general execution after training?
16:18 < fenn> also seems like a good starter product for scaling up DNA barcoding for connectomics
16:18 < fenn> biological stuff dies and then you have to re-train it. also you can't work out the bugs on one organoid and transfer that knowledge to another. it doesn't scale
16:19 < kanzure> you can use machine learning and genome synthesis (or genome editing) to grind towards a genome that produces a biological neural network that once trained has more legible connection weights
16:19 < fenn> i also want a pony
16:19 < kanzure> such as by labeling the connection weights
16:19 < kanzure> ok then you are talking about something else..?
16:19 < fenn> you mean like brainbow stuff?
16:20 < fenn> literally colorful synapse weights?
16:20 < kanzure> whatever; we don't even know if receptor density measurements are what we need to replicate biological execution in hardware- but with a tight feedback loop you can just have the computers figure it out by modifying the biology until there's a signal that they are monitoring and using for feedback
16:20 < fenn> how do you know you have successfully emulated the bio network?
16:21 < kanzure> the simulated or emulated version matches the behavior of the physical one (physical reservoire computing stuff goes here)
16:21 < kanzure> just faster in silico
16:21 < fenn> not necessarily faster
16:22 < kanzure> you've lost me. so not for performance but for longevity you want weight transferance from in vivo to in silico?
16:22 < kanzure> s/longevity/replicability
16:23 < fenn> silicon is more parallelizable i guess; for large networks on current GPU architectures with memory separate from compute, the cost of moving bits around swamps everything else and even a 20Hz brain is faster
16:24 < fenn> it costs about the same to run 1 mind as 16 minds, and you get diminishing returns around a batch size of 64
16:24 < kanzure> what do you mean "even though" there's no direct readout of network weights? how did you get to "distill their output logits" if there's no weight readout?
16:25 < fenn> a blob of flesh on an electrode array with N electrodes has N "output neurons" which we can compare to the input and output layers of a neural network
16:26 < fenn> if we train the flesh to output probabilities of the next word, say, that information can be used to efficiently train a neural network to think the same way
16:26 < fenn> we aren't looking at it under a microscope at all
16:27 < fenn> this technique is being used already for transferring knowledge between language models and image generation models
16:27 < fenn> er, between models of the same type, including such types as language and image
16:27 < kanzure> idea is biological tissue would learn faster/more cheaply? because training is expensive compute/energy when in silico?
16:27 < fenn> right
16:28 < kanzure> and transferance is through repeated querying? i haven't been following the LLM stealing prompt stuff
16:28 < fenn> also maybe we can replicate the thinking pattern by training another blob of flesh to reproduce the probability distributions
16:29 < kanzure> (the thing where you query an LLM a bunch of times and reconstruct its weights in another model)
16:29 < fenn> going directly from flesh to flesh might work, i have to think about it
16:29 < fenn> yeah that is what i am talking about with "distillation"
16:29 < kanzure> what is the term of art for that?
16:29 < fenn> it's called that because usually you go from a large model to a small model
16:30 < fenn> "model extraction"
16:33 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has joined #hplusroadmap
16:55 < L29Ah> https://en.wikipedia.org/wiki/James_Magnussen#2024
17:02 -!- darsie [~darsie@84-113-55-200.cable.dynamic.surfer.at] has quit [Ping timeout: 260 seconds]
17:04 -!- BEXCHA is now known as autopilot
17:04 < fenn> "In February 2024, Magnussen announced that he would be coming out of retirement to take part in the Enhanced Games. Magnussen will be paid $1 million if he breaks the men’s 50m freestyle record"
17:10 < fenn> is there some kind of mutual snitching attractor that socially excludes athletes who talk about enhancement? 
17:11 < fenn> is it just wikipedia's editor bias? surely there must be someone willing to speak in favor of the enhanced games
17:14 < fenn> growth hormone mutant? you're the lucky winner! taking growth hormones? that's unfair and immoral~
17:17 < fenn> either way you have to be absurdly dedicated to actually win
17:17 < fenn> the way i see it, freedom of substrate actually levels the playing field. no more genetic lottery
17:18 < fenn> i prefer unlimited engineering competitions like pikes peak vs over-regulated f1 or god forbid nascar
17:19 < fenn> otherwise what are we even doing? starting out with the conclusion, then performatively executing the foregone conclusion
17:23 < fenn> in 1968 a creative engineering team swapped out their old piston engine for a gas turbine helicopter engine. it was so good it immediately won all the races and was banned. we're now stuck with the piston engine technology in this supposedly elite but actually second rate league
17:25 < fenn> in 1978 another creative team used a fan to generate extra downforce, keeping the car stuck to the track even at low speeds. the car immediately won all the races and was banned. we're now stuck with passive aerodynamics
17:26 < L29Ah> same thing with cycling
17:26 < L29Ah> there are devices that reach 130kph on a horizontal track by human power alone
17:28 < L29Ah> one could then argue for allowing/banning energy capacitors tho
17:29 < fenn> ok but surely there should at least be one sports league that allows all the silly things
17:30 < fenn> like omg what if they allowed turbochargers on race cars, then all cars would have to use turbochargers!
17:30 < fenn> only naturally aspirated engines are safe
17:30 < fenn> the turbine might explode and kill someone with high speed shrapnel
17:30 < fenn> see how stupid that sounds
17:30 < fenn> but the line is completely arbitrary
17:32 < fenn> WADA's decisions on which substances to allow are not based on risk, and they don't even pretend to quantify the risk of "natural" athletics
17:33 < fenn> the stuff modern athletes do is so far removed from even 50 years ago, that it wouldn't be a fair competition if you could time travel
17:33 < fenn> it's not "just" that there's a larger pool to select from
17:34 < fenn> science has progressed, but some science has been banned
17:44 < kanzure> fenn: the silicon adherents would argue that in silico training is faster than biological neural network training and therefore your in vitro tissue or organoid culture scheme would not be advantageous
17:46 < kanzure> makes sense that pro-transhumanism athletes would be the ones that have already retired from the regular athletic careers and have no concerns about losing out on other competitions by participating in pro-steroid sports
17:49 < fenn> participating in enhanced games would be career suicide for any current pro athlete
17:49 < fenn> this is literally a rebellion against the IOC
17:49 < kanzure> does WADA have a chokehold on pro skating?
17:50 < fenn> no
17:50 < kanzure> but WADA infiltrated chess
17:50 < kanzure> can't explain that
17:50 < fenn> when did that happen
17:51 < fenn> skaters are much more distrustful of authority and drug testing than most cultures
17:51 < kanzure> https://www.fide.com/FIDE/handbook/WADA%20Anti%20Doping.pdf
17:51 < kanzure> https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5978818
17:51 < kanzure> "According to the Chess WADA Anti-Doping Policy, the most relevant banned substances for chess are amphetamine derivatives (Adderall, Ritalin), ephedrine and methylephedrine, pseudoephedrine, and Modafinil. Notably, caffeine and codeine are not prohibited, but figure in the Monitoring Program."
17:52 < kanzure> looooool
17:52 < kanzure> machines are crusihng our best chess players and we're busy trying to cap our best chess players by monitoring their caffeine intake
17:53 < kanzure> "look these coffee beans natively have dextroamphetamine okay?"
17:54 < fenn> i can't even find anything about doping in skatboarding
17:55 < kanzure> olympic skating is subject to WADA.. makes sense.
17:55 < kanzure> "With skateboarding's inclusion in the Olympics starting from Tokyo 2020, athletes competing in these events must comply with WADA's anti-doping rules. This means undergoing drug testing and adhering to the list of prohibited substances. Olympic skateboarding competitions are governed by the International Skateboarding Federation (ISF) or World Skate, which have adopted anti-doping policies in ...
17:55 < kanzure> ...line with WADA's standards."
17:56 < kanzure> "Professional skateboarding has historically been characterized by its countercultural ethos, which includes a relaxed attitude towards substance use."
17:56 < kanzure> guess it didn't totally escape WADA.
17:56 < fenn> yes but world skate is all kinds of different activities involving wheels or blades
17:57 < kanzure> "... organized by specific communities or organizations dedicated to extreme pogo or "Xpogo," and these organizations may not have strict anti-doping policies in place"
17:57 < fenn> also japanese skaters have very different culture
17:59 < kanzure> perhaps a privacy-focused enhancement olympics could work, where you keep everything private and anonymous
17:59 < fenn> lol yeah right
17:59 < fenn> homomorphically encrypted athletics competitions
17:59 < kanzure> "and here's the 100m guy fawkes dash"
18:00 < fenn> "DO YOU NOT HAVE A MOLE ON YOUR LEFT ANKLE?"
18:01 < hprmbridge> kanzure>  https://cdn.discordapp.com/attachments/1064664282450628710/1205694945407533136/Ftm8phPWABUbz5h.png?ex=65d94deb&is=65c6d8eb&hm=eb226253b3f33e3e086dccc3391a070a61948a713c112b29ec774cc02e684702&
18:01 < fenn> i would be interested in whatever substances skaters have found improves dexterity, balance, coordination, timing, etc.
18:01 < fenn> it's mostly not about strength or endurance
18:02 < fenn> also there's some cognitive capability involving trajectories through state space, which i don't know how to name
18:02 < kanzure> hopefully it's not completely bruteforced muscle memory 
18:03 < fenn> it's not
18:03 < fenn> practice is key of course, but some people just learn to skate faster than others
18:05 < kanzure> what argument is there that biological training would be faster? i think i saw something like amount of data exposure in common crawl (or whatever data sets they are using for 150b models?) is several orders of magnitude larger than human child exposure
18:05 < kanzure> but, human child exposure is massive just from retinal input... so...
18:05 < fenn> baby chickens learn at about the same rate as visual transformers
18:05 < fenn> uh, sample efficiency
18:07 < fenn> this whole thing is relevant https://www.lesswrong.com/posts/xwBuoE9p8GE7RAuhd/brain-efficiency-much-more-than-you-wanted-to-know
18:08 < fenn> it's not that biology is inherently superior, it's that the architecture of current silicon designs is not suited to the task
18:08 < fenn> you need memory dispersed within the parallel compute nodes, the more dispersed the better
18:08 < fenn> biology takes it to the limit where the memory IS the compute
18:09 < fenn> a 1:1 allocation
18:09 < kanzure> hm, this link does not talk about learning rates
18:10 < kanzure> ok maybe under vision
18:10 < fenn> https://arxiv.org/abs/2312.02843 baby chicken sample efficiency vs vision transformer
18:11 < fenn> spoiler: they're the same within an order of magnitude
18:11 < kanzure> .wik human speechome project
18:11 < saxo> "The Human Speechome Project ('speechome' as an approximate rhyme for 'genome') is an effort to closely observe and model the language acquisition of a child over the first three years of life. /  The project was conducted at the Massachusetts Institute of Technology's Media [...]" - https://en.wikipedia.org/wiki/Human_speechome_project
18:11 < EmmyNoether> "The Human Speechome Project ('speechome' as an approximate rhyme for 'genome') is an effort to closely observe and model the language acquisition of a child over the first three years of life. /  The project was conducted at the Massachusetts Institute of Technology's Media [...]" - https://en.wikipedia.org/wiki/Human_speechome_project
18:11 < kanzure> nsh: bots are fighting with each other again
18:12 < kanzure> "and transcribing significant speech added a labor-intensive dimension" this needs a whisper update
18:13 < kanzure> something something podcast speech anonymization filters
18:14 < fenn> baby babble is not going to work with whisper
18:14 < kanzure> "The resulting corpus, which already contains over 100,000 hours of multi-track recordings, constitutes the most comprehensive record of a child's development made to date."
18:14 < fenn> wait, they only have a single data point?
18:14 < fenn> wtf
18:14 < fenn> literally anyone could do this at any time
18:15 < kanzure> no follow-up publications since 2012 https://www.media.mit.edu/cogmac/projects/hsp.html
18:15 < fenn> strap a cellphone to a baby. done
18:15 < fenn> jesus christ what are the other 8 billion people doing
18:15 < kanzure> so parents already do that
18:16 < kanzure> you could probably just take an average across a few million children that are (already) placed in front of phones all day
18:16 < fenn> what makes "the human speechome" special then?
18:16 < kanzure> oh, AFAIK this is the only project that actually did it
18:16 < fenn> the cellphone has to be turned on and recording audio, of course
18:16 < kanzure> i just mean that you don't have to convince anyone to strap a smartphone to a baby--- they do it naturally (thus the debates about screen time)
18:17 < fenn> presumably this must be done earlier than when babies are given devices
18:17 < fenn> like a newborn can't even move in a coordinated fashion, what's it going to do with a phone
18:17 < kanzure> i suspect the american pediatric association recommends no screen time before 2 years of age specifically because parents were giving children screen time before 2 years of age
18:18 < fenn> they had to be seen giving some sort of recommendation
18:18 < kanzure> sorta dark i guess...
18:18 < fenn> the alternative is colorful things hanging from strings over the crib
18:19 < kanzure> kinda weird that this speechome project doesn't even have a stats page
18:19 < kanzure> how many hours of actual speech?
18:19 < kanzure> how many words were exposed? what was the entropy? what was the word learning rate for this subject?
18:20 < kanzure> and then compare learning rate and language acquisition rate against LLMs or something (or whisper?)
18:22 < fenn> that's basically what they did with the chickens
18:24 < fenn> i think the data quality and ... curriculum preparation (if there even is one) is much worse for a typical 2021 LLM than for a human 
18:24 < kanzure> i don't think human curriculums are likely well organized, it's probably 99% just some variation of immersion learning
18:24 < kanzure> for first 1-2 years
18:24 < fenn> they're treating it like a big blender when there is evidence that LLM learn faster from simplified synthetic data sets
18:25 < kanzure> okay but nobody has figured out simplified human synthetic data sets AFAIK
18:25 < kanzure> it's like making contact with an alien species and you want to transmit prime numbers except the party on the other end of the line isn't cognitively developed yet
18:26 < fenn> so obviously the next step is presenting the training data in a structured order right? but somehow the sequel to "tiny stories" (synthetic data) was not "tiny computer school" but rather "let's generate random python code examples"
18:26 < kanzure> i forget what this is called. there was some scheme someone devised. you start by transmitting numbers and primes, and then make up arithmetic from there, and then you introduce words somehow. was this from a movie or was it a real thing attached to the VGER probe?
18:27 < fenn> contact had something like that. the NSA had a decryption challenge along those lines, and it's been done to death in science fiction
18:27 < fenn> the pioneer record had some math stuff iirc
18:27 < fenn> the story of your life
18:28 < kanzure> SETI's messaging program (METI) isn't the right fit because there's no interaction between the two parties, it's just one way transmission
18:28 < fenn> grocery store's about to close, i'll bbiab
18:29 < kanzure> "Algorithmic communication systems are a relatively new field within CETI. In these systems, which build upon early work on mathematical languages, the sender describes a small set of mathematic and logic symbols that form the basis for a rudimentary programming language that the recipient can run on a virtual machine."
18:29 < kanzure> that doesn't make sense. why would the recipient have a compatible virtual machine?
18:30 < kanzure> "Logic Gate Matrices (a.k.a. LGM), developed by Brian McConnell, describes a universal virtual machine that is constructed by connecting coordinates in an n-dimensional space via mathematics and logic operations, for example: (1,0,0) <-- (OR (0,0,1) (0,0,2)). Using this method, one may describe an arbitrarily complex computing substrate as well as the instructions to be executed on it."
18:31 < kanzure> "Astrolinguistics: Design of a Linguistic System for Interstellar Communication Based on Logic" or "lingua cosmica astrolinguistics"
18:32 < kanzure> .wik astrolinguistics
18:32 < saxo> "Astrolinguistics is a field of linguistics connected with the search for extraterrestrial intelligence (SETI). /  Arguably the first attempt to construct a language for interplanetary communication was the AO language created by the anarchist philosopher Wolf Gordin (brother [...]" - https://en.wikipedia.org/wiki/Astrolinguistics
18:32 < EmmyNoether> "Astrolinguistics is a field of linguistics connected with the search for extraterrestrial intelligence (SETI). /  Arguably the first attempt to construct a language for interplanetary communication was the AO language created by the anarchist philosopher Wolf Gordin (brother [...]" - https://en.wikipedia.org/wiki/Astrolinguistics
18:33 < kanzure> https://cosmicos.github.io/
18:36 < fenn> .wik solomonoff induction
18:36 < saxo> "Solomonoff's theory of inductive inference is a mathematical theory of induction introduced by Ray Solomonoff, based on probability theory and theoretical computer science." - https://en.wikipedia.org/wiki/Solomonoff_induction
18:36 < EmmyNoether> "Solomonoff's theory of inductive inference is a mathematical theory of induction introduced by Ray Solomonoff, based on probability theory and theoretical computer science." - https://en.wikipedia.org/wiki/Solomonoff_induction
18:38 < kanzure> a variation of this work that allows for two-way interactive communication (eg testing and dynamic feedback adjustment) would be interesting and could be applied to cross-language human communication, human-animal communication, and machine learning model extraction
18:39 < kanzure> "CosmicOS is a way to create messages suitable for communication across large gulfs of time and space. It is inspired by Hans Freudenthal's language, Lincos, and Carl Sagan's book, Contact. CosmicOS, at its core, is a programming language, capable of expressing simulations. Simulations are a way to talk, by anology, about the real thing they model."
18:42 < kanzure> unfortunately this (cosmicos) requires too much abstract study by the recipient. language immersion training is usually highly interactive and has feedback and does not require logical analysis or careful intentional mental simulation.
19:44 < fenn> it turns out i had an extra hour before they closed
19:44 < fenn> me grunting about solomonoff induction was attempting to point at a method by which alien intelligences could arrive at a minimal shared virtual machine
19:45 < fenn> and not be overfit to the data in the transmission
19:47 < fenn> we share not only the transmitted bits themselves, but also the rest of the observable universe. that ought to narrow down the search space of compatible virtual machines by quite a lot
19:49 < fenn> if we're being honest, even without any special effort on the sender's part, any alien intelligence worth talking to ought to be able to figure out the meaning of most symbols from context and shared observation of natural phenomena
19:49 < fenn> math language is baby-speak for aliens i guess
19:55 < fenn> i would be cautious trying to use this for anything beyond bootstrapping, because it's not going to be a system widely used and tested by humans and probably contains a lot of bugs and stupid mistakes just due to being an obscure topic without much history (not enough eyes to find the shallow bugs)
20:12 < fenn> kinda surprised examine.com has nothing about shiitake
20:16 < hprmbridge> Eli> Examine doesn’t have enough people helping out. Even their well documented stuff is missing a lot of papers. Somehow they need to automate the process
20:17 < fenn> gentlemen, we have the technology
20:18 < fenn> examine.com is a for-profit company that doesn't accept outside contributions
20:18 < fenn> i like the concept; i might even be willing to pay for the service. someone needs to be doing quality control of supplements and claims about supplements, and i don't trust the FDA with this task
20:19 < fenn> just an independent "certified not bullshit" stamp
20:19 < fenn> but in general there just isn't enough research to draw from in the first place, and nobody is willing to fund it
20:20 < fenn> maybe we could tax clinical trials, since they already are throwing zillions of dollars at them
20:20 < fenn> logically it makes sense to find new uses for substances that have already been proven to be safe, thus bypassing the need for safety trials
20:21 < fenn> i don't see a need for efficacy trials before a drug can be distributed
20:47 -!- mxz [~mxz@user/mxz] has quit [Ping timeout: 272 seconds]
20:56 -!- Gooberpatrol66 [~Gooberpat@user/gooberpatrol66] has joined #hplusroadmap
21:00 -!- Ashstar [~Ashstar@mobile-166-171-251-240.mycingular.net] has joined #hplusroadmap
21:00 < Ashstar> https://link.springer.com/article/10.1007/s11357-023-00936-w
21:01 < Ashstar>  Blood biomarker profiles and exceptional longevity: comparison of centenarians and non-centenarians in a 35-year follow-up of the Swedish AMORIS cohort | GeroScience
22:52 -!- mxz [~mxz@user/mxz] has joined #hplusroadmap
--- Log closed Sat Feb 10 00:00:54 2024