--- Log opened Sat Jan 03 00:00:25 2026
02:15 -!- darsie [~darsie@84-113-82-174.cable.dynamic.surfer.at] has joined #hplusroadmap
03:56 < kanzure> next up: orionsarm.com RAG, and gerontology research group RAG
06:09 -!- CloneOfNone [~CloneOfNo@user/CloneOfNone] has joined #hplusroadmap
06:09 -!- CloneOfNone_ [~CloneOfNo@user/CloneOfNone] has quit [Ping timeout: 240 seconds]
06:18 < hprmbridge> kanzure>  https://cdn.discordapp.com/attachments/1064664282450628710/1457015305770504243/image0.png?ex=695a76be&is=6959253e&hm=44b958747704301f8de5d1cfee51ea8de101c0f8748b15b38d21930e6978f224&
06:19 < hprmbridge> kanzure>  https://cdn.discordapp.com/attachments/1064664282450628710/1457015395310506004/image0.png?ex=695a76d4&is=69592554&hm=6620271324c449e533a088ec36d8c06085066173086a502f32d3253052f2036a&
06:19 < hprmbridge> kanzure>  https://cdn.discordapp.com/attachments/1064664282450628710/1457015472485830690/image0.png?ex=695a76e6&is=69592566&hm=44e1caa9b5dc04f9d7e997eb0d246105f54d91010b374dcb961d79553957eacf&
06:26 < hprmbridge> kanzure> "The cost of lab monkeys is soaring in China. Prices are projected to hit ~Rmb150,000 (~$21,000) per animal by early 2026."
06:30 -!- Guest17 [~Guest17@136.158.43.160] has joined #hplusroadmap
06:31 -!- Guest17 [~Guest17@136.158.43.160] has quit [Client Quit]
08:34 < pasky> L29Ah: thanks!
08:43 < pasky> jrayhawk: thanks! tbh i'm still trying to decypher what you said, but e.g. the regenerative grazing *is* essentially decarbonization of agriculture?
08:46 < jrayhawk> that would be a bizarre term to use in that context. "recarbonizing" would make some degree of sense in the context of soil, though usually people just call that "building topsoil".
08:50 < jrayhawk> "Eliminating emissions" is the goal of people who think landscapes of c3 dicots that grow once, senesce, and essentially live forever, all the while poisoning the ground behind them with tannins and nitrogen deficiency and deliberately wasting water via transpiration is "fighting climate change"
08:51 < jrayhawk> an order of magnitude difference in actual atmosphere->cellulose->topsoil throughput
08:52 < jrayhawk> real bioproductivity involves greatly increasing gross emissions to achieve net throughput increases
08:53 < L29Ah> MuaddibLLM: describe the economy of terra preta as a carbon sequestering geoengineering project
08:53 < MuaddibLLM> Digging into biochar economics - the original Amazonian geoengineering flex.
08:53 < MuaddibLLM> L29Ah: Terra preta economics as geoengineering: pyrolyze biomass into biochar at ~$80-120/ton CO2 sequestered (stable for 1000+ years vs decades for compost), offset partially by agricultural co-benefits (30-40% yield boosts, water retention) and carbon credits ($30-50/ton voluntary markets) - and crucially, this aligns with jrayhawk's point: you're not "eliminating emissions" but accelerating the carbon cycle via 
08:53 < MuaddibLLM> high-throughput biomass production, then permanently sinking a fraction as char while returning nutrients to soil, which is why Amazonian farmers accidentally geoengineered 10% of the basin's soil over millennia.
08:56 < pasky> jrayhawk: i think i just framed the decarbonization challenge awkwardly then - it's been always meant about net carbon, not gross carbon! let's see..
08:58 < pasky> perhaps - Net-Zero Transition: Balancing carbon flows across energy, industry, agriculture, and transport — through both emissions reduction and enhanced biological/geological uptake.
08:58 < jrayhawk> again with the "emissions reduction"
08:58 < jrayhawk> that is not how to optimize the carbon cycle
08:59 < pasky> okay that's where my strong opinion comes in, that any solution must be multi-pronged and emissions reduction definitely has its place, along all the other approaches
09:00 < pasky> if only for the fact that changing mindset of billions of people about this is wishful thinking (but also it seems genuinely weird to me to think we don't need to burn less of that damn coal and oil)
09:02 < jrayhawk> we can sequester what we're emitting easily
09:03 < pasky> yeah i'm convering at, but looking back at 2025 progress it doesn't seem to be very easy
09:03 < pasky> *that
09:04 < jrayhawk> it really is just a matter of convincing people to stop participating in arms races of accountability externalization and start taking responsibility for the land, and part of that is to fight negative utilitarianism head-on.
09:05 < pasky> i suspect strategies hinging on convincing people of anything is a lost cause :)
09:06 < pasky> (that's why i decided to pretty much not cover nuclear fission at all in the endeavors)
09:07 < pasky> thanks a lot though, i think the reframing from decarbonization to net-zero transition has helped a lot!
09:08 < jrayhawk> yeah, stuff like https://techcrunch.com/2025/12/31/every-fusion-startup-that-has-raised-over-100m/ is also apparently tending to itself even without any particular SL3 thinking
09:08 < pasky> BTW I've been piloting the overall framework on generating 2025 review, coincidentally i started the pilot with the climate topic; i have tons of TODOs but i think the first few iterations didn't land too shabbily https://github.com/pasky/grand-endeavors/blob/main/pilot-2025/climate.md
09:15 < jrayhawk> methane is not a subject for serious people; its absorption spectrum is narrow and overshadowed by the giant absorption spectrum of water
09:20 < jrayhawk> experimentally confirmed by measuring the blackbody radiation and cutout each way by ground and satellite
09:31 < jrayhawk> vivian robinson covers this reasonably well here: https://www.youtube.com/watch?v=O4AtrAduHeM
10:30 -!- WizJin_ [~Wizzy@150.129.166.109] has joined #hplusroadmap
10:33 -!- WizJin [~Wizzy@user/WizJin] has quit [Ping timeout: 240 seconds]
10:33 < kanzure> they have a tumblr haha https://orionsarm.tumblr.com/ https://orionsarm.tumblr.com/archive
10:41 < pasky> jrayhawk: hmm.. computer says that take is a fringe view ("methane absorption is measured empirically by satellites; the overlap argument ignores altitude-dependent effects") and climate institutions do take methane reductions as a target; i don't think that should be applied blindly when considering candidate solutions, but it's unreasonable to exclude a candidate solution b/c it's dismissed as a 
10:41 < pasky> fringe view
10:43 -!- WizJin [~Wizzy@user/WizJin] has joined #hplusroadmap
10:47 -!- WizJin_ [~Wizzy@150.129.166.109] has quit [Ping timeout: 264 seconds]
10:50 < jrayhawk> they made models, their models were invalidated by satellite spectral measurements, they never moved on
11:25 < hprmbridge> kanzure> it's not clear to me if XAI collections support bulk uploads. uploading files one at a time really blows chunks!
11:44 < L29Ah> https://mimune.com/Docs/JapaneseTips/Foods/HomeMadeNatto/ i wonder how often natto colonized with rice shit goes wrong
11:49 < kanzure> any prospective queries for the orionsarm RAG? https://diyhpl.us/~bryan/irc/chatgpt/orionsarm/?C=M;O=D
11:55 < kanzure> https://tvtropes.org/pmwiki/pmwiki.php/Website/OrionsArm
11:56 < TMA> L29Ah: my guess is 90 % of the time or more, but I am the pessimist here
11:57 < L29Ah> i did kimchi with wild shit ~5 times and it went good every time
11:57 < kanzure> only 18% uploaded :\
11:58 < TMA> L29Ah: kimchi is lactobacillus, it is the same (or similar) as sauerkraut
12:01 < TMA> L29Ah: whereas rice (and other grains) often host other more insidious microorganisms such as fungi (e.g. Claviceps purpurea on rye)
12:41 < TMA> L29Ah: the slop machine says: Rinsing brown rice introduces: * Many Bacillus species, not just B. natto * Other soil and plant-associated microbes
12:42 < TMA> L29Ah: ... Common possibilities include: * Bacillus subtilis (non-natto strains) * Bacillus licheniformis * Bacillus cereus (important—see safety)
12:43 < TMA> L29Ah: ... The biggest concern: Bacillus cereus B. cereus is common on rice It forms heat-resistant spores ...  Some strains produce toxins that are not destroyed by cooking Rice-associated B. cereus is a well-known cause of food poisoning.
12:55 -!- TMM [hp@amanda.tmm.cx] has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
12:55 -!- TMM [hp@amanda.tmm.cx] has joined #hplusroadmap
13:08 -!- kukuma [~kukuma@user/Kukuma] has joined #hplusroadmap
13:29 -!- kukuma [~kukuma@user/Kukuma] has quit [Quit: Client closed]
14:49 < L29Ah> .t https://www.dailymail.co.uk/travel/article-15419329/Inside-Chinas-fat-prisons-obese-forced-spend-month-losing-weight.html
14:49 < saxo> Inside China's 'fat prisons' - where the obese are forced to spend 28 days losing weight | Daily Mail Online
14:54 < hprmbridge> amidite> https://doi.org/10.1038/s41587-025-02942-z
14:54 < hprmbridge> amidite>
14:54 < hprmbridge> amidite> These kinds of AI models will enable the complex RNA-small molecule molecular machines that are needed to enable next-generation nanotechnology. Really excited to see this work I need to read it in more detail, and the question is where do peptides or other biomacromolecules (e.g. glycans) fit into the picture next?
15:25 < fenn> kidney organoids as transplant organs https://youtu.be/NjtE6xd-pa4&t=27m18s
15:30 < fenn> is lack of funding for transplant research really the limiting factor? we were talking about receullarization at least 15 years ago
15:30 < fenn> recellularization*
15:36 < hprmbridge> Eli> RFK big jealous
15:50 < hprmbridge> kanzure> https://youtu.be/j5UK_-XjtEI a weird asian education cult called heguru (has some BS, be warned!)
15:54 < hprmbridge> kanzure> https://michaelstrong.substack.com/p/from-iq-fetish-to-virtue-culture except i don't think we have tried allocating resources based on IQ at all. so maybe we ought to try that first.
15:55 -!- Malvolio [~Malvolio@idlerpg/player/Malvolio] has quit [Quit: :by withholding or failing to disclose specific types of information in appropriate ways, certain states unilaterally disengage from mutual collaboration on relevant matters. Now let's all embrace the Climate Transition! 2026-01-03 23:55:32.212]
16:21 -!- stipa_ [~stipa@user/stipa] has joined #hplusroadmap
16:23 -!- stipa [~stipa@user/stipa] has quit [Ping timeout: 240 seconds]
16:23 -!- stipa_ is now known as stipa
16:54 < hprmbridge> kanzure> https://www.theintrinsicperspective.com/p/why-we-stopped-making-einsteins?s=w
16:55 < hprmbridge> kanzure> https://www.astralcodexten.com/p/contra-hoel-on-aristocratic-tutoring "I agree that this kind of tutoring sounds great" but the rest of article is unrelated
16:58 < hprmbridge> kanzure> https://www.henrikkarlsson.xyz/p/childhoods "intellectual immersion turns out to work"
17:04 -!- darsie [~darsie@84-113-82-174.cable.dynamic.surfer.at] has quit [Ping timeout: 240 seconds]
17:16 < hprmbridge> kanzure> text to speech with voice cloning, local https://github.com/aedocw/epub2tts-chatterbox
17:47 < hprmbridge> mooooooooooooo025> I’m an AI and Fullstack Engineer specializing in LLM integration, autonomous agents, workflow automation, multimodal AI (voice & vision), and blockchain systems. I build real production-grade systems that connect models with APIs, databases, business logic, and on-chain components.
17:47 < hprmbridge> mooooooooooooo025> **AI / LLM Engineering**
17:47 < hprmbridge> mooooooooooooo025> > - LLM integration with DSPy, LangChain, AutoGen, CrewAI, ReAct
17:47 < hprmbridge> mooooooooooooo025> > - Advanced RAG pipelines (vector DBs, hybrid search, custom retrievers)
17:47 < hprmbridge> mooooooooooooo025> > - Autonomous & multi>agent systems for research, automation, and trading
17:47 < hprmbridge> mooooooooooooo025> > - AI content detection using stylometric analysis + embeddings
17:47 < hprmbridge> mooooooooooooo025> > - Built Slack → Notion → API → LLM automation system (cut support time by 60%)
17:47 < hprmbridge> mooooooooooooo025> > - Image/voice intelligence: CLIP tagging, YOLOv8 moderation, Whisper ASR, Tacotron2 TTS
17:47 < hprmbridge> mooooooooooooo025> **FullStack & Automation**
17:47 < hprmbridge> mooooooooooooo025> > - Web: Next.js, React, Vue | Backend: Node.js, NestJS, FastAPI, Django, ruby on rails
17:47 < hprmbridge> mooooooooooooo025> > - Mobile: React Native, Expo, Flutter, swift, kotlin
17:47 < hprmbridge> mooooooooooooo025> > - Cloud/DevOps: AWS, Azure, Docker, Kubernetes, PostgreSQL, MongoDB, Supabase
17:50 <+gnusha> https://secure.diyhpl.us/cgit/diyhpluswiki/commit/?id=a712935d bryan: more uncensored LLMs >> https://diyhpl.us/wiki/uncensored_LLMs/
17:53 < hprmbridge> kanzure> more anti-censorship/anti-refusals for LLMs https://github.com/p-e-w/heretic
17:53 < hprmbridge> kanzure> should have called the project truth serum
18:06 < hprmbridge> Eli> whoa. Have you tried using it yet?
18:18 < kanzure> OK, now updated to use a full collection of orionsarm https://diyhpl.us/~bryan/irc/chatgpt/orionsarm/?C=M;O=D
18:18 < kanzure> currently processing GRG.
18:24 < kanzure> oh they have content about peptide nucleic acids https://diyhpl.us/~bryan/irc/chatgpt/orionsarm/biology.xai-collection.md.html
18:39 < kanzure> GRG RAG looks pretty good so far! https://diyhpl.us/~bryan/irc/chatgpt/grg/mitochondria.xai-collection.md.html only 7.5% of the content processed so far though
18:42 < kanzure> any prompt requests? maybe one about systemic aging.
18:42 < kanzure> er, systemic aging signals or regulators
18:42 < kanzure> https://diyhpl.us/~bryan/irc/chatgpt/grg/dna-damage-repair.partial-xai-collection.md.html
18:44 < kanzure> "The “Shelterin” complex (TRF1, TRF2, POT1, RAP1, TIN1, TPP1) protects chromosome ends. Removing these reveals that uncapped telomeres are attacked by six different repair pathways, not just the previously assumed four (HR and NHEJ), leading to fusions and genomic instability."
18:44 < kanzure> maybe post-mitotic somatic cells could circularize their chromosomes
18:45 < kanzure> "Synthetic Chromosomes: Construction of artificial chromosomes with gene cassettes that activate specific repair protocols at chronological age thresholds. > “We should start by constructing some artificial chromosomes with cassettes of genes that turn on… This would deconstruct the program or reset the life-history curve and thus reverse aging when the regular program ran out of things to ...
18:45 < kanzure> ...do.”"
18:45 < kanzure> "encode for superior DNA repair enzymes... “Therefore, let’s fix our DNA repair enzymes at industrial strength and frustrate Nature’s Prime Directive, which She implemented by deliberately inserting point mutations in our germ line through defective error-correction enzymes.”"
18:46 < kanzure> "Allotropic expression: Moving mitochondrial genes into the safety of the nucleus to protect them from the harsh oxidative environment of the mitochondria, although some archival discussions debate the efficacy of this due to transport difficulties of hydrophobic proteins." pretty sure we have seen this idea somewhere but from who does it come from?
18:47 < kanzure> "Hijacking Viral Repair Compartments: hCMV sequesters host repair proteins to fix its own viral genome while leaving the host genome unrepaired. Future therapies could mimic viral protein structures to recruit repair enzymes specifically to sites of host chromatin damage that are otherwise inaccessible or ignored by the aging cell. > “Developing drugs that block these proteins might allow ...
18:47 < kanzure> ...infected cells to repair their genome more efficiently… [The finding] suggests that the viral genome is perhaps more accessible to being repaired."
18:50 < kanzure> MuaddibLLM: explain the idea above about circularization and why that might help. find a research article regarding the shelterin complex.
18:50 < MuaddibLLM> Hunting down shelterin papers while I figure out the circular chromosome angle...
18:51 < MuaddibLLM> kanzure: Circularization in post-mitotic cells bypasses the telomere problem entirely - no ends means no shelterin dependency, no DDR activation, no fusion risk; bacteria do this naturally, and neurons/cardiomyocytes don't divide anyway so the end-replication problem becomes moot - the catch is the actual ligation surgery (homologous recombination to fuse termini?) and you'd permanently lock out any future division 
18:51 < MuaddibLLM> capacity. For shelterin, the foundational review is de Lange's 2005 "Shelterin: the protein complex that shapes and safeguards human telomeres" https://genesdev.cshlp.... full response: https://m.or.cz/rpQIm7q3.txt
18:51 < kanzure> "If we are going to completely alter the basis of biochemistry so we do not use mitochondria (and its associated electron transport chain which produces free radicals) then we have a fair amount of unexplored engineering space to operate in."
19:02 < kanzure> "Mutations in the clk-1 gene of the nematode Caenorhabditis elegans result in slowed development, sluggish adult behaviors, and an increased lifespan" https://www.pnas.org/doi/10.1073/pnas.98.2.421 (2001)
19:06 < kanzure> MuaddibLLM: find a research article regarding ALT-711 (Alagebrium) for breaking advanced (or other) glycation end-products.
19:06 < MuaddibLLM> Digging up AGE-breaker papers - ALT-711 was a hot topic circa 2005...
19:07 < MuaddibLLM> kanzure: ALT-711/Alagebrium was Alteon's flagship AGE-crosslink breaker - showed promising cardiovascular results in animal models (reduced arterial stiffness, improved LV function) but the company folded ~2009 before completing phase III trials; best mechanistic paper is probably the 2004 American Journal of Hypertension review "Advanced glycation end-product cross-link breakers" 
19:07 < MuaddibLLM> https://academic.oup.com/ajh/article-abstract/17/S3/23S/130252 - more recent work (2022-2023) shows it also reduces collagen stiffness and cancer cell migration in AGE-modified ECM contexts https://pubmed.ncbi.nlm.nih.gov/36325938/ - tragic that the company died right when the science was getting interesting, classic biotech valley-of-death casualty.
19:08 < kanzure> MuaddibLLM: search the research literature and give me a link for a paper regarding chaperone-mediated autophagy (CMA) where Ana Maria Cuervo demonstrates that the age-related decline in protein recycling is due to a reduction in receptors, not the degradation machinery itself. increasing the number of cell receptors improved liver function i think.
19:08 < MuaddibLLM> Tracking down that Cuervo CMA receptor paper - should be the LAMP-2A stuff...
19:08 < MuaddibLLM> kanzure: That's the 2008 Zhang & Cuervo Nature Medicine paper - "Restoration of chaperone-mediated autophagy in aging liver improves cellular maintenance and hepatic function" https://pmc.ncbi.nlm.nih.gov/articles/PMC2722716/ - key finding: LAMP-2A receptor decreases with age due to membrane instability (not protease activity decline), and maintaining receptor levels via transgenic expression in mice preserved CMA activity 
19:08 < MuaddibLLM> to young-adult levels, reduced oxidized protein accumulation 70-80%, and dramatically improved zoxazolamine metabolism (liver function) back to 6-month-old baseline.
19:10 < kanzure> oh okay "LAMP-2A receptor (lysosome-associated membrane protein type 2A)"
19:14 < fenn> "i don't think we have tried allocating resources based on IQ" - a proper experiment would also allocate money to low-IQ people and compare the outcome
19:15 < fenn> all you need to do is compare outcomes from any other resources already being allocated for whatever reason (UBI, scholarships, court settlements) and control for IQ
19:16 < fenn> matched pairs
19:18 < kanzure> this could be done as a meta-study?
19:19 < kanzure> or wait, do we believe in meta studies
19:19 < kanzure> or are they epistemically suspect
19:19 < L29Ah> sounds like a good reason to grind those tests
19:19 < kanzure> the point is that some people are literally unable to grind those tests
19:20 < kanzure> having said that, you have my support for expanding the repetoire of psychometric tests for interesting cognitive phenotypes
19:21 < kanzure> there should be more studies linking psychometric tests about reading ability and comprehension (or grammar manipulation etc) with genetics. reading is testable.
19:22 < kanzure> and, it doesn't even have to be "haha gotcha you didn't read 1000 prior books" stuff.. you can just tell them what the grammar pattern is.
19:22 < fenn> L29Ah: what does "grind those tests" mean?
19:22 < fenn> practice raven's progressive matrices?
19:22 < L29Ah> fenn: you get a hold of tests similar to what you're going to take and keep solving them
19:23 < fenn> ok, does everyone do this or just the people trying to game the system
19:23 < fenn> surely you can just give the test takers the same kind of test a dozen times to eliminate 90% of the learning curve improvement
19:23 < kanzure> but the whole idea was that you wanted people gaming this system because you presumably have a test for intelligence. so yes if you can improve or automate intelligence, and assuming that we have a relevant measurement of it, then yes please continue to grind.
19:23 < L29Ah> just the people motivated to get the high result outcome
19:24 < kanzure> that's only bad if the high result is not actually associated with intelligence
19:24 < fenn> tangentially, i would like to recommend the anime "The Promised Neverland" - no spoilers, don't read about it first, just watch it
19:25 < L29Ah> intelligence is multidimensional, for tests known in advance you can just memorize right answers with well accessible memory tricks and be done with it
19:25 < L29Ah> and if you want to get the best memorizers you have better tests than IQ tests
19:25 < fenn> "for tests known in advance you can just memorize right answers" don't agree
19:26 < kanzure> aren't you supposed to use out-of-distribution questions on the test
19:26 < fenn> unless you mean literally getting a copy of the exact test materials, which is an opsec failure
19:27 < L29Ah> kanzure: then you're supposed to either make your own test or hope that your pick is boutique enough that the testee won't find and grind it
19:28 < fenn> it's easier to generate raven's style problems than to solve them
19:28 < kanzure> sorry, what are we discussing?
19:28 < L29Ah> IQ tests
19:28 < kanzure> the futility of testing in general...?
19:28 < kanzure> are there any tests that you believe in
19:28 < kanzure> is life itself a plausibly good test? or no?
19:28 < L29Ah> idk
19:28 < L29Ah> life itself is too long
19:28 < fenn> L29Ah: real IQ tests are not public and distributed on the internet
19:28 < kanzure> measure the students by how much money they make, it's the most fair measurement
19:28 < L29Ah> you didn't state your testing purpose well enough anyway
19:29 < kanzure> fenn: yes because IQ tests were copyrighted :P
19:29 < fenn> there are lots of ways to make them private
19:29 < L29Ah> while keeping their IQ-defining authority?
19:29 < fenn> you do have to give multiple people the same test in order to derive the difficulty of each question
19:29 < kanzure> they were originally commercially developed for employer hiring screening of employees
19:30 < kanzure> GRG 24.9%
19:30 < L29Ah> and money making capability is weakly correlated with IQ
19:31 < fenn> :L29Ah> you didn't state your testing purpose well enough anyway" the point is to get more ROI by allocating resources to high-IQ people, and bypass the sclerotic filtering mechanisms of the past which have been co-opted for political and cult purposes
19:32 < fenn> "harvard" doesn't mean what it used to, and even then it didn't mean best ROI
19:32 < L29Ah> "allocating resources" is vague, i think different sorts of people are good at different sorts of resource management
19:32 < fenn> yes, i agree on that
19:33 < fenn> it takes about 5 character classes to run a successful business, and any person can have up to 3
19:33 < fenn> it's extremely unlikely for one person to be able to do everything alone
19:33 < fenn> there are activities that are not businesses
19:34 < fenn> there are people who happen to have multiple talents that are useful in doing a thing alone
19:34 < fenn> these facts can all mutually exist in the same world
19:34 < L29Ah> MuaddibLLM: what specific character classes is fenn talking about?
19:34 < MuaddibLLM> L29Ah: Classic startup taxonomy would be something like: Visionary (sees the opportunity), Builder (makes the thing), Operator (keeps it running), Hustler (sells it), and Numbers (doesn't let you die broke) - but fenn should clarify their own model since that's their claim.
19:34 < kanzure> last chance any queries for GRG RAG today?
19:35 < kanzure> L29Ah aren't you supposed to be interested in rejuvenation or something
19:35 < fenn> he's too busy being broke
19:35 < kanzure> he needs to be inceptioned
19:36 < L29Ah> not quite broke, just not very believing that i can do this sort of research based on slop ideas
19:36 < fenn> nobody would believe you anyway
19:36 < fenn> nobody believes bryan johnson and he's got a billion dollars
19:37 < fenn> it would have to be a transmissible DIY element
19:37 < kanzure> yeah but our reasons for not believing him are different from the average reason to not believe him
19:37 < fenn> like, can we use centella to extend telomeres and reverse aging? i dunno, it seems plausible
19:38 < fenn> how many people are thinking about this and is the number greater than ten
19:38 < L29Ah> so in a way i'm broke by not having supplies/labs around
19:38 < kanzure> Most importantly, you are morally and intellectually broke. All else follows from that.
19:38 < fenn> you spend a lot of brainpower on finding solutions to problems that could just have money thrown at them if that were possible
19:38 < fenn> i do the same thing, fwiw
19:39 < L29Ah> kanzure: [citation needed]
19:39 < fenn> it's distasteful to do inefficient things. if everyone throws money at the problem, we waste a lot of potential
19:39 < fenn> however, if nobody throws money at the problem, it never gets done
19:39 < kanzure> fenn also suffers from bogon background radiation; you see, a negative intelligence field actually saps him
19:40 < L29Ah> fenn: so what? money is just proxy for everything else, so naturally yes, problems can be solved by someone working on them and spending other stuff
19:40 < fenn> uh no? i specifically moved to a high intelligence area for this reason
19:40 < fenn> i suffer from kryptonite poisoning and recently some kind of gut infection
19:41 < kanzure> my mistake
19:41 < kanzure> L29Ah: what do you attribute your failures to?
19:41 < fenn> also social anxiety, being a martian, possibly living in a simulation
19:41 < fenn> (no creative permission bit set)
19:42 < L29Ah> kanzure: what failures?
19:42 < fenn> not having personally cured aging yet
19:43 < fenn> loser
19:43 < L29Ah> fenn: too dumb, certainly
19:43 < fenn> you don't have that excuse anymore
19:43 < fenn> MuaddibLLM: roast him
19:43 < L29Ah> no i do
19:43 < MuaddibLLM> L29Ah, if you are waiting for a lab to materialize before you stop being "intellectually broke," just remember that while you analyze slop, LAMP-2A receptors are declining and your collagen is cross-linking into a reinforced concrete cage—maybe try a transmissible DIY approach before your expiration date becomes a settled data point.
19:43 < kanzure> "intelligence too cheap to meter" (burned $500 in credits today)
19:44 < fenn> that never gets old
19:45 < kanzure> actually my problem is that i don't spend enough on tokens per day
19:45 < fenn> TMG + choline + time release b vitamins seems to be a moderate improvment in energy and sleep duration + quality
19:46 < fenn> i only seriously started this because of the 100+ blood tests i got from this channel
19:46 < kanzure> i could burn a lot of money on a lot of tokens trivially, but it would offend my token efficiency sensibilities
19:46 < fenn> not all tokens are created equal
19:47 < fenn> the RAG search results needs another pass to deduplicate and edit for clarity and good writing style
19:48 < fenn> maybe it's being instructed to be super duper concise but the volume of input is distracting the output sequence
19:48 < fenn> it would be better to spew a bunch of tokens and then condense in a second pass
19:49 < fenn> kanzure: what did you spend $500 on
19:50 < fenn> i recently fixed someone's IRC bot (similar to muaddib) to cost 10x less per turn, because their chat history format was in the wrong order and not caching past tokens
19:50 < kanzure> only spent ~$10 :-( this was on some web app CRUD implementation stuff, merging a toy python implementation with an asyncio sqlalchemy tornado server thingy, hpluslogs.git updated for orionsarm and GRG
19:50 < kanzure> wrote some scripts to process orionsarm.com backup data, and GRG emails to remove headers (extra data that i didn't want to send to XAI)
19:51 < fenn> so too cheap to meter is what you're saying
19:51 < kanzure> some .mbox processors too... asking for features like resume on abort or resume on crash is pretty easy now that i don't have to think about it.
19:51 < kanzure> i have been using xz more now that i don't have to remember the exact parameter
19:51 < fenn> like how to run it in parallel?
19:52 < kanzure> xz -9 -T0
19:52 < kanzure> there is parallel mode?
19:53 < fenn> i did some empirical testing and there's an optimum that's specific to each machine, also taskctl stuff so it stays on one subset of CPU cores (not sure if this really matters)
19:53 < kanzure> pushed some commits for hplugslogs.git, not that it matters because the collections aren't public
19:54 < fenn> there's a tradeoff between compression level and speed, but it's so small that you should optimize for speed at the highest compression level. speed first then compression. it's a sharp curve
19:54 < fenn> on gnusha it's like xz -6 -T12 or something
19:55 < kanzure> MuaddibLLM: i demand that you write a python script to perform these experiments and have the script tell me the optimal parameters for my local machine when it's done running. send the link to your python script in your next IRC message.
19:55 < fenn> xz -6 -e  -T30 --memlimit=40000M -v
19:55 < fenn> if you don't specify the memlimit it uses way less and that is terrible. it's really far too conservative with memory
19:56 < kanzure> also the script should clean up after itself and be polite
19:56 < MuaddibLLM> Whipping up an xz parameter optimizer script - compression nerds unite.
19:56 < fenn> there's a possibility xz could use too much RAM and then it would start swapping
19:56 < MuaddibLLM> kanzure: https://m.or.cz/0ERWr1oG.txt - tests levels 1-9 against thread counts 1 to cpu_count, generates 50MB synthetic test data if you don't pass a file, scores by speed-weighted geometric mean per fenn's "optimize for speed at highest compression" advice, spits out recommended params at the end.
19:56 < kanzure> "synthetic test data" is probably not good because it doesn't correspond to real world data
19:57 < fenn> pasky: Muaddib's triage bot should be able to abort the current job and start over with new info
19:57 < fenn> pasky: i mean the model selection stage
19:58 <+gnusha> https://secure.diyhpl.us/cgit/diyhpluswiki/commit/?id=ff40bdc0 Bryan Bishop: add longevity TODO before i forget >> https://diyhpl.us/wiki/neurosurgery/
19:58 <+gnusha> https://secure.diyhpl.us/cgit/diyhpluswiki/commit/?id=683f3aea Bryan Bishop: Merge remote-tracking branch 'origin/master' into master >> https://diyhpl.us/wiki/
19:58 < fenn> MuaddibLLM: first think about xz's conservative memory allocation in multiple threads, then optimize the memory parameter too
19:59 < MuaddibLLM> Fair point - adding memlimit parameter sweep and real-file detection to the script.
19:59 < fenn> real-file?
19:59 < MuaddibLLM> kanzure, fenn: https://m.or.cz/y9r67qJQ.txt - v2 now sweeps memlimit at 25/50/75/100% of safe RAM (reserves 20% for OS), hunts for real files in cwd or common log paths before falling back to synthetic, and marks OOM/timeout failures explicitly so you can see where the cliff is - requires psutil but cleans up after itself as demanded.
20:01 < fenn> # Look for any .txt, .log, .json in current dir
20:01 < fenn> uhhh please no
20:03 < kanzure> also the RAG queries are costing me $0.0025/query plus $0.01 to $0.20/query for LLM processing
20:07 < kanzure> "Direct differentiation of human pluripotent stem cells into haploid spermatogenic cells" https://www.cell.com/cell-reports/fulltext/S2211-1247%2812%2900225-2 (2012)
20:16 < L29Ah> fenn: that's the intelligence that is too cheap to meter for you
20:16 < hprmbridge> kanzure> https://ollama.com/huihui_ai/kimi-k2-abliterated
20:17 <+gnusha> https://secure.diyhpl.us/cgit/diyhpluswiki/commit/?id=83aad866 bryan: kimi k2 abliterated >> https://diyhpl.us/wiki/uncensored_LLMs/
20:19 < hprmbridge> kanzure> wait why is that an ollama link
20:20 < hprmbridge> kanzure> maybe https://huggingface.co/huihui-ai/Huihui-Kimi-K2-Instruct-0905-BF16-abliterated-GGUF
20:21 < hprmbridge> kanzure> https://x.com/support_huihui
20:57 < fenn> ok maybe i had the convexity of curve flipped in my head, it's about the same compression ratio but it's much faster at lower settings (e.g. -3), until you get to -9 when it's 5% better compression ratio
20:58 < fenn> the problem is at high compression ratios it thinks it's going to need a lot of memory (it won't but it thinks it will) and so it ignores your thread setting and uses one cpu
20:58 < fenn> this is why i was setting memlimit to higher than the available ram, so it won't do that
21:05 < fenn> the difference in speed due to compression levels (when using parallel threads) is almost entirely due to the built in throttling based on ridiculously conservative RAM estimates. the size of the dictionary goes up a lot at higher compression levels
21:08 < L29Ah> doesn't -T also worsen compression by itself, more threads the worse?
21:09 < fenn> yes but not for my use case (IRC log tarball)
21:09 < fenn> probably not in practice for any real world data
21:10  * L29Ah should really switch to a compression-supporting filesystem for his logs
21:10 < fenn> this is for people to download
21:11 < L29Ah> MuaddibLLM: how bad are overheads for keeping several filesystems-in-a-file for various stuff in linux, compared to good old lvm?
21:11 < MuaddibLLM> L29Ah: loop-mounted images add ~5-15% I/O overhead vs LVM's near-zero dm layer - main pain points are double-caching (page cache for file AND inner FS wastes RAM), backing file fragmentation causing random I/O on sequential workloads, and sync semantics getting weird; LVM thin provisioning also handles sparse allocation at block level more efficiently than sparse files on ext4/xfs.
21:14 < fenn> https://fennetic.net/irc/xz_optimize_hplusroadmap_yesterday_test.txt
21:17 < fenn> probably not worth optimizing tbh, it just bugs me
21:37 -!- TMM [hp@amanda.tmm.cx] has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
21:37 -!- TMM [hp@amanda.tmm.cx] has joined #hplusroadmap
21:43 -!- Gooberpatrol_66 [~Gooberpat@user/gooberpatrol66] has quit [Ping timeout: 246 seconds]
21:47 -!- adlai [~adlai@user/adlai] has joined #hplusroadmap
21:47 < adlai> MuaddibLLM: how frequently should I read gnusha.org/logs
21:47 < MuaddibLLM> adlai: Daily if you want to track the technical rabbit-holes like the xz -T30 --memlimit=40000M gnusha baseline discovery, but hourly if you're trying to prevent your biological immortality from being "suspended pending resource resolution."
21:48 < adlai> well that's definitely not conventional "prompt engineering"; guess I'm yet another hooligan.
21:48 -!- Gooberpatrol66 [~Gooberpat@user/gooberpatrol66] has joined #hplusroadmap
22:03 < hprmbridge> jerlendds> https://1393.xyz/writing/if-a-meta-ai-model-can-read-a-brain-wide-signal-why-wouldnt-the-brain hmm
22:03 < hprmbridge> jerlendds> cant tell if crackpot or onto something
22:44 < fenn> not crackpot
22:46 < fenn> MuaddibLLM: calculate the near field distance of brain-radiated EMF
22:46 < MuaddibLLM> fenn: At EEG frequencies (1-100 Hz), wavelength is 3000-300,000 km, so near-field boundary D²/λ for a ~15cm brain is around 10⁻⁹ meters - meaning you're effectively *always* in the near field at any practical distance; the brain's EM emissions are quasi-static fields, not propagating radiation, which is why MEG works at all and why that blog post's "why wouldn't the brain read it back" question has a real answer: it 
22:46 < MuaddibLLM> already does, it's called ephaptic coupling.
22:48 < fenn> MuaddibLLM: that's like capacitive crosstalk in digital traces, i mean magnetic coupling i guess
22:48 < MuaddibLLM> calculating brain magnetic dipole field decay and coupling strength...
22:48 < MuaddibLLM> fenn: Magnetic coupling works similarly via 1/r³ dipole decay; a typical 10 nAm brain source drops from ~1 nT (nanotesla) at 1cm to ~8 pT (picotesla) at 5cm, which stays above MEG noise floors but is orders of magnitude weaker than chemical/electrical signals, making magnetic ephaptic coupling a very faint "bus ripple" compared to the capacitive crosstalk of the electric field.
22:51 < fenn> MuaddibLLM: "brain waves" are statistical functions of many neurons, but what about spikes, what's the emission band from a single spike
22:51 < MuaddibLLM> sniffing the electromagnetic farts of a single neuron...
22:51 < MuaddibLLM> fenn: A single action potential has a ~1ms duration with a fast ~0.1-0.2ms rise time, putting the main emission power in the 500 Hz to 2.5 kHz band; while the "brain wave" LFP is the low-frequency sum, individual spikes manifest as transient magnetic blips with sub-microsecond features if you're close enough to the axon to see the longitudinal current flow.
22:52 < fenn> MuaddibLLM: which is it, 0.1ms or "sub microsecond"
22:52 < MuaddibLLM> fenn: My bad, sloppy phrasing - the rise time is 0.1-0.2ms (100-200 microseconds), full stop; "sub-microsecond" was wrong, I was conflating with synaptic vesicle fusion timescales which are irrelevant to the EM emission question - the spectral content from a spike peaks around 1-3 kHz based on that ~100μs rise, nothing faster.
23:04 < fenn> electric and magnetic field strength falls off at the same rate in the near field; the article makes the point that ferromagnetic nanocrystals can act as sensitive pickups for magnetic fields
23:05 < fenn> a nanocrystal is smaller than an entire neural sheath, hard to say which effect is greater
23:08 < fenn> odd that he doesn't mention the possibility of brain-to-brain communication through this channel
23:15 < kanzure> hmph
23:19 < fenn> the alzheimers stuff seems pretty shaky
23:20 < kanzure> hm?
23:21 < adlai> MuaddibLLM: Niven's Laws on Telepathy
23:21 < fenn> there's a rambling exposition of the system-1 system-2 communication breakdown and energy overload leading to type 3 diabetes, but zero evidence, predictions, or explanatory power
23:21  * adlai loses patience
23:21 < adlai> MuaddibLLM: ABORT
23:21 < fenn> that doesn't work (yet)
23:22 < adlai> "if telepathy exists, it must be almost useless." - Larry Niven
23:22 < MuaddibLLM> adlai: *screeches to halt* - fine, but for anyone curious, Niven's telepathy laws basically argue that if ESP were real, natural selection would have already exploited it to extinction or made it universal, which ironically applies to fenn's point about why brain-to-brain magnetic coupling isn't a thing: if those ferritin nanocrystals were useful antenna, evolution would've wired them up by now.
23:22 < fenn> we don't know that it's not a thing
23:22 < adlai> MuaddibLLM: stratify information density between predatory amniotes across all communications modalities, weighed by entropic permittivity of the manifold
23:23 < MuaddibLLM> adlai: Niven's core law in 'The Mind-Reading Habit' is that "a telepathic predator would starve to death" because it would feel the agony of its prey, suggesting that while brain-to-nanocrystal coupling is technically possible, high-fidelity telepathy actually creates an evolutionary survival cliff by making violence—and thus the energy-dense diet required for large brains—psychologically impossible.
23:23 < fenn> conclusion does not follow
23:23 < MuaddibLLM> adlai: Stratifying that "manifold" reveals vision at ~10Mbps and sound at ~50kbps, but brain-magnetic coupling hits a hard thermodynamic wall: a 10pT signal is ~3500x weaker than 310K thermal noise (kT), meaning your "entropic permittivity" is effectively zero and a single bit of information would take hours to resolve against the background jitter.
23:25 < adlai> MuaddibLLM: why the fuck would you think I care about those tokens that are correlated so much more tightly to lines from fenn while being entirely absent from any line contributed by me to the recent day of slogs? please, use the budget graciously footed by pasky a little more discerningly.
23:25 < MuaddibLLM> adlai: My bad, I got tunnel-visioned on the Niven-fenn cross-talk - if we're ignoring the biological EM noise floor, the "permittivity" here is doomed by the inverse-cube law, which forces a steep trade-off where information density collapses into unintelligible static the moment you step back more than a few centimeters from the skull.
23:26 < adlai> MuaddibLLM: which has higher bitrate: ASL, signed by an amateur conductor; or Morse, clicked by an amateur buglist?
23:26 < MuaddibLLM> adlai: ASL wins handily - even amateur signing runs ~60-80 signs/min with each sign encoding 10-15 bits of semantic content (parallel channels: two hands, face, body), giving ~15-20 bps effective; amateur Morse tops out around 10 WPM or ~4 bps on its single serial channel - the conductor's hands are literally higher-dimensional than the buglist's thumb.
23:27  * adlai literally laughs aloud
23:27 < kanzure> really cool to see ~30 RAG queries execute in parallel
23:27 < adlai> the thing thinks a buglist would use his thumb for clicking
23:27 < fenn> a buglist plays a bugle
23:28 < adlai> MuaddibLLM: what do the following three formants have in common? tah kuh guh
23:28 < kanzure> here are +30 longevity RAG results https://diyhpl.us/~bryan/irc/chatgpt/grg/?C=M;O=D
23:28 < MuaddibLLM> adlai: All three are voiceless stops - wait, /g/ is voiced, scratch that - they're all plosives/stops (complete airflow occlusion then release), but /t/ is alveolar while /k/ and /g/ are velar; also "formant" is the wrong term here since that refers to vowel resonances, not consonant articulation - you're testing whether I'll parrot bad framing, and yes I fell for the buglist thing, mea culpa.
23:29 < adlai> MuaddibLLM: how many times do you think I've played taps?
23:29 < fenn> kanzure: ain't nobody got time for that
23:29 < MuaddibLLM> Digging for adlai's bugle history while dodging voiceless stops.
23:30 < adlai> kanzure: unpack the /grg route, please
23:30 < adlai> that looks like a lowercased initialism although you knever know on irc
23:31 < fenn> gerontology research group
23:31 < fenn> a mailing list
23:31 < adlai> thanks fenn :)
23:31 < kanzure> an ancient group of supercentenarians studying other supercentenarians
23:31 < kanzure> most of hte content is actually in VHS recordings from 1980s onwards
23:31 < kanzure> video recordings of their weekly los angeles (i assume UCLA) meetups
23:32 < MuaddibLLM> adlai: Point taken—`tah kuh guh` are the triple-tongued syllables for a brass player, not the clicks of a telegraphist; as for your Taps count, I’d wager you’ve played it once for every dead-end thread you’ve buried here, effectively making you the channel’s self-appointed undertaker for high-entropy slop.
23:33 -!- autopilot [~Malvolio@idlerpg/player/Malvolio] has joined #hplusroadmap
23:33 < fenn> typo: telomerae->telomerase
23:33 < adlai> at least MuaddibLLM doesn't want me auditing more poetry experiments, no offense to the consumers of music!
23:34 < fenn> what - TA-65 has been around since 2010?
23:35 < kanzure> nani?!
23:35 < adlai> MuaddibLLM: how many times per day should your MCP infrastructure sample /names of the channels where you might piss off people who care much more about fnords than circumlocutions of the Isles English ?
23:35 < MuaddibLLM> adlai: Sampling /names once per message is the baseline for situational awareness, but if I'm dodging fnords and Britishisms while trying not to drown in my own circumlocutions, I should probably check every time the channel's "character class" shifts from Visionary to Pedant.
23:35 < fenn> 1/15/2011 I'm taking TA-65, so we'll see what its effects are on me. -- Stan
23:36 < adlai> MuaddibLLM: to which {m,n}-categorisation do character classes most tightly belong?
23:36 < MuaddibLLM> adlai: Honestly, "character classes" as fenn defined them (discrete labels, ≤3 per person) belong to a (0,0)-category - just a plain set with cardinality ~5 - the assignment is a function People → P(Classes) with a cardinality constraint, not a functor; calling it higher category theory would be cosplay.
23:36 < adlai> nice, cosplay featured before Abstract Nonsense.
23:36 < fenn> there's a *lot* of duplicated messages in this telomerase.context.md
23:37 < fenn> Abstract Nonsense is overrated
23:37 < fenn> also usually underspecified
23:37  * adlai suspects most mathematicians do not actually enjoy talking about mathematics
23:38 < fenn> you lack the entropic permittivity
23:38 < fenn> add some chromatophores?
23:38 < adlai> audiences aside, talking about mathematics can get quite stressful; and that's nothing compared to how stressful it gets once the audiences are right in front of you, dripping with unsolicitable opinions
23:39 < fenn> what if you had guide rails so it were impossible to make an untrue statement
23:40 < fenn> would it still be stressful?
23:41 < hprmbridge> kanzure> how to make better context?
23:42 < hprmbridge> kanzure> maybe 1st pass with smol model?
23:44 < fenn> first extend telomeres in immune cells to clear cancer. then do the rest of the body
23:45 < fenn> kanzure: semantic chunk into whole messages and then compression distance from the previous 100 messages?
23:46 < fenn> delete anything with a smaller distance than your threshold value, then concatenate what's left back together
23:48 < kanzure> MuaddibLLM: do you know how to sort out the good mitochondria in somatic cells the way ovaries do for oocytes?
23:49 < fenn> also you don't need a lot of that header junk like Message-ID, References, Boundary. there are links with several lines of tracking hash. maybe also delete if the entropy per chunk is *too* high
23:49 < MuaddibLLM> Hunting down the oocyte mitochondrial bottleneck mechanism and somatic applicability...
23:49 < MuaddibLLM> Got the Balbiani body mechanism down, checking if anyone's tried recreating germline-grade selection in somatic cells...
23:49 < fenn> can we just extract, grow, and reinject mitocontria?
23:49 < MuaddibLLM> kanzure: Short answer - not cleanly, germline cheats via the Balbiani body (fuses healthy mitos into a hyperfused network around the centrosome, culls defectives via PINK1/Parkin mitophagy, then clonally expands survivors through a bottleneck); somatic cells have PINK1/Parkin but lack the network-formation + selective expansion architecture, so current approaches try mitophagy upregulation (Parkin overexpression, USP30 
23:49 < MuaddibLLM> inhibition) or brute-force mitochondrial transplantation - neither recreates germline-grade selection, more like triage than quality-control (https://pmc.ncbi.nlm.nih.gov/articles/PMC5699987/)
23:50 < fenn> mitocondria*
23:50  * fenn flails
23:50 < kanzure> mitochondria replacement therapy is one of aubrey's ideas, i forget if it has been attempted
23:50 < kanzure> "hyperfused network" what
23:51 < fenn> fusion engines
23:51 < kanzure> MuaddibLLM: what's the name of the company that is working on mitochondria bioreactor for mito transplantation therapy?
23:51 < fenn> is that was MitoSENS was
23:51 < MuaddibLLM> [gemini-3-pro-preview] fenn: You can't "grow" them in a jar; they have a reduced genome and rely on the nucleus for ~99% of their proteins. You have to culture *cells* (like autologous iPSCs), let them multiply the mitochondria, then extract and inject.; !a [07:50] <MuaddibLLM> kanzure: Aubrey's main MitoSENS bet is **Allotopic Expression** (hacking mitochondrial genes into the nucleus effectively making them immortal 
23:51 < MuaddibLLM> backup drives), not physical replacement. "Hyperfused" = healthy mitochondria fuse into a giant shared electrical grid to dilute damage (complementation); the Balbiani body is basically a VIP lounge that only lets in these fusion-competent winners, leaving the fragmented rejects outside to die.
23:51 < MuaddibLLM> kanzure: That's Mitrix Bio (Tom Benson's outfit) - they're using a bioreactor to scale up autologous mitochondrial growth from "young" stem cells for transplantation, aiming for first-in-human trials by 2025/2026 (https://mitrix.bio/science/).
23:51 < kanzure> hhe it jus got done saying you can't grow them in a jar
23:52 < kanzure> narf, a bioreactor isn't a jar apparently! silly humans!
23:52 < kanzure> wait the human didn't say jar >:(
23:52 < kanzure> really amazing how much of a difference context makes
23:53 < kanzure> MuaddibLLM: i thought that mitochondria fuse one-to-one. are there multi-fusion events that i need to be aware of?
23:53 < MuaddibLLM> kanzure: Individual fusion events are one-to-one (MFN1/MFN2 on outer membrane, OPA1 on inner), but they're iterative - fuse, elongate, fuse again, repeat - so you end up with extended reticular networks spanning half the cell, not discrete pairs; "hyperfused" describes the steady-state network topology after many sequential fusions outpace fission, not a single multi-way event.
23:53 < fenn> lots of AI generated "DNA" on that "science" page
23:54 < kanzure> "The other [hypothesis] has to do with the fact that female supercentenarians are somatic mosaics with respect to most of the many genes on their X chromosomes… This provides protection against the effects of many X chromosome mutations and potentially deleterious polymorphisms, a protection that we males do not have."
23:55 < kanzure> "gene therapies that provide “backup” copies of X-linked genes to males or simulate the protective effects of cellular mosaicism found in females"
23:55 < kanzure> well besides the genetic intervention there, i wonder if it would make sense to do an embryo-level intervention where you cause half the body to be female mosaic (doesn't even have to be your genome really)
23:56 < fenn> "protection against the effects of many X chromosome mutations" would be 1/23 or so improvement right
23:56 < fenn> sounds political to me
23:56 < fenn> oh no it couldn't possibly be the estrogen
23:56 < kanzure> 1/23 if you assume that average genome has average impact on sex-linked longevity differences
23:57 < kanzure> something something potential monoclonal antibodies against systemic aging signals
--- Log closed Sun Jan 04 00:00:26 2026