--- Log opened Wed Aug 06 00:00:02 2025
01:48 < fenn> yeah gpt-oss is probably not worth bothering with, the "safety" team got to it
02:24 < fenn> this is just bot abuse https://i.redd.it/uud2hotmubhf1.png
02:33 < jrayhawk> surgery is expensive and dangerous for everyone, but it's especially dangerous for someone immunologically and metabolically broken
02:44 < jrayhawk> while there is a positive feedback cycle between macrophages releasing IL-6 while cleaning up leaky adipocytes and IL-6 inducing hypothalamic leptin resistance that causes upward drift in weight over time, that cytokine load is still minority influence vs. other inputs most of the time
02:51 -!- TMM [hp@amanda.tmm.cx] has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
02:51 -!- TMM [hp@amanda.tmm.cx] has joined #hplusroadmap
02:52 < jrayhawk> In the feedback endgame of the leaks overcoming endogenous cleanup capacity causing feet to rot off, I can see the cytokine load becoming majority influence, but even that's a situation where mixed {gamma,alpha}-toco{pherols,trienols} makes more sense to stabilize the situation than liposunction.
02:53 < jrayhawk> i mean the adipocyte content becoming directly causal of the majority of the cytokine load
03:09 < hprmbridge> .monokhrome> ozempic claimed to have anti-aging effects: https://trial.medpath.com/news/5c43f09ebb6d0f8e/ozempic-shows-anti-aging-effects-in-first-clinical-trial-reversing-biological-age-by-3-1-years
03:12 < jrayhawk> The more wholistic solution is 1) decrease PUFA ratio closer to where can be utilized for synthesis/structure (long-chain), or OxPhosed, rather than stored 2) alleviate other immunological burdens to decrease cytokines to decrease hypothalamic leptin resistance to decrease appetite to decrease overall calorie storage, and 3) improve OxPhos capacity with mitochondriogenesis signals, growth 
03:12 < jrayhawk> substrates, and electron transport chain cofactors.
03:13 < fenn> decrease PUFA ratio of diet or of adipocyte composition or what?
03:14 < jrayhawk> the adipocyte composition follows from the diet
03:14 < jrayhawk> so, yes
03:14 < fenn> yes but there's an "assload" of it there that must be dealt with first
03:15 < fenn> i worry that freeing up large quantities of stored PUFA will cause problems
03:17 < jrayhawk> There are circumstances where it can. Genetic lipoprotein processing disorders, for instance, exacerbate the fact that lipoproteins are non-metabolic and can't do antioxidant recycling or membrane repair.
03:18 < jrayhawk> This is a place where PCSK9 inhibitors make sense.
03:20 < jrayhawk> Rapid fat mass loss is also notorious for releasing other accidentally bioaccumulated fat soluable toxins.
03:21 < jrayhawk> Which is why a lot of studies measuring health parameters try to maintain weight
03:22 < fenn> and something about ketone bodies increasing oxidative stress
03:24 < jrayhawk> this gets very funny in situations like Lynda Frassetto's paleo trials where she was trying to study kidney disease and having to feed people 5000 calories per day and discourage them from suddenly spontaneously exercising in order to maintain weight.
03:25 < fenn> i've heard the paleo diet described as "eat expensive food" - is there a way to do it that won't bankrupt the planet and/or convert everything into cow infrastructure?
03:27 < fenn> if you feed chickens corn won't they just uptake the corn oil and store it as PUFA?
03:27 < fenn> i know pigs do when fed peanuts
03:30 < jrayhawk> I don't understand the paleo question. Nothing is more expensive than cleaning up after the fact with medical care, which is the alternative to mesa-alignment. Basically anything is more ecologically sustainable than standard western dieting.
03:31 < fenn> USDA says 1 standard chicken thigh contains 10g of fat, saturated 2.8g, mono 4.0g, poly 2.2g, (mostly 18:2 omega-6)
03:32 < fenn> perhaps on an individual level you're right but when more and more land is devoted per person the cost will increase as demand outstrips supply
03:33 < jrayhawk> And yeah, for most people (that don't have lipoprotein processing disorders), birds have mediocre lipid ratios compared to pork, and pork has mediocre lipid ratios compared to cows, but all that's still nothing compared to peanut-oil-fried doughnuts and frenchfries.
03:39 < hprmbridge> .monokhrome> I remember my geology professor claiming that fat loss would release heavy metals and toxins. She was obviously practiced what she preached, as she was fat.
03:40 < jrayhawk> haha
03:40 < fenn> No food named "LARVA".  Try another (or <enter> to quit)
03:40  * fenn grumbles
03:42 < jrayhawk> fenn: you're thinking like a californian again. the quality of life of every last thing on this planet is defined in terms of throughput on the carbon cycle. the most bioeffective systems are C4 monocot-based, and humans can't directly metabolize the majority of c4 monocot biomass. (Sedge tubers we can do okay on.)
03:44 < fenn> i just want a fusion reactor outputting tasty ice cream okay?
03:44 < hprmbridge> .monokhrome> what flavor?
03:44 < fenn> algae flavor
03:44 < jrayhawk> you get to manage the land to keep it productive (i.e. preventing senescence and monopolization) either with fire or grazing
03:44 < jrayhawk> 80% of arable land is already marginal land only suitable for grazing
03:45 < jrayhawk> C4 monocot agronomy is the only good means we have of restoring failed ecosystems
03:46 < fenn> i could think of a few alternatives
03:46 < fenn> actually restoring the praries, actually having large herds of buffalo and wolves etc
03:47 < jrayhawk> what energy pathway do you think the praries used?
03:48 < fenn> you know that's not how it's going to go
03:49 < jrayhawk> https://en.wikipedia.org/wiki/Native_American_use_of_fire_in_ecosystems this is how we got those praries
03:49 < fenn> yes
03:49 < jrayhawk> all that released energy could've gone towards ruminants instead and achieved the same ends
03:50 < jrayhawk> they just have to be actively managed to do so
03:55 < fenn> how about hydrogenating with better catalysts that need lower temperatures and so won't form trans fats
03:56 < fenn> i mean this is predicated on food scientists knowing the downstream consequences of what they're doing, which i don't have a lot of hope for, but maybe AI will help guide them
03:56 < jrayhawk> well, some of those transfats turn out to be quite useful, such as conjugated linoleic acid and vaccinic acid
03:56 < jrayhawk> which is why cows produce them
03:58 < fenn> this is... trans omega-7?
03:58 < fenn> the terminology in this field sucks
04:02 < fenn> can we just add BHT to everything?
04:03 < fenn> "generally recognized as safe"
04:04 < fenn> "BHT stops this autocatalytic reaction by converting peroxy radicals to hydroperoxides."
04:05 < jrayhawk> Stopping propagation, not initiation.
04:06 < fenn> the idea would be to load up your fatty tissues with enough BHT to bring oxidation down to what it would be if you had a mostly saturated fat diet
04:06 < jrayhawk> Tocopherols and tocotrienols are already doing that job.
04:07 < jrayhawk> Stopping propagation is useful, but it is not a complete solution.
04:09 < fenn> can we just feed pigs/chickens high-oleic sunflower oil instead of soy/corn/peanut
04:10 < fenn> i don't understand why they would even grow the other varieties
04:10 < fenn> it just goes bad faster
04:11 < jrayhawk> I don't know the specific history of sunflowers, but the unholy alliance of Nixon and McGovern led to a lot of very bad policy around lipid ratios.
04:11 < fenn> tocopherol dosage: "Plasma F2-isoprostane concentration was selected as a biomarker of free radical-mediated lipid peroxidation. Only the two highest doses - 1600 and 3200 IU/day - significantly lowered F2-isoprostane"
04:12 < fenn> "1 IU is also defined as 1 milligram of an equal mix of the eight stereoisomers"
04:12 < fenn> oof
04:13 < fenn> 100g of sunflower oil contains 61IU
04:14 < jrayhawk> Yeah, toco{pherol,trienol}-to-PUFA ratio is the thing to optimize for
04:14 < hprmbridge> .monokhrome> Humans and our pre-human animal ancestors must have gone through their lives gaining and losing weight - how can evolution have left the process so harmful?
04:15 < jrayhawk> In the evolutionary training distribution, we weren't eating and storing pounds of soybean oil.
04:16 < fenn> it's not the gaining and losing weight that's harmful, it's the rancid fat
04:17 < jrayhawk> There is a lot of reason to believe humans are programmed to gorge on seaonally available fruit, but excess calories from that is converted primarily into palmitic acid, which is stable and safe to store.
04:18 < fenn> there are some kinds of fat that we can smell when it goes rancid, and it smells disgusting. those fats have have been carefully removed by a process of bleaching and refinement
04:18 < hprmbridge> .monokhrome> is palmitic acid the one in chocolate?
04:18 < fenn> palmitic acid is 16:0 a medium-long saturated fatty acid
04:20 < fenn> "It is derived from palm oil, palm kernel oil, coconut oil and milk fat" it's also present in chocolate but not especially noteworthily so
04:20 < fenn> maybe you're thinking phenethylamine
04:21 < fenn> i kinda wish we would just dump all this nomenclature and use numbers to describe fatty acids
04:21 < fenn> it's not that complicated
04:21 < jrayhawk> https://minio.scielo.br/documentstore/1678-457X/n5SQd8KgwzfZxsfKDfKtZKS/d8247db14f68ca1b44ee67944996546c21e661d8.jpg
04:22 < fenn> is that the chocolate spectrum
04:23 < jrayhawk> yeah
04:23 < hprmbridge> .monokhrome> well google says palmitic acid is found in large amounts in both milk fat and cocoa butter, so it's a large part of chocolate
04:23 < jrayhawk> or, rather, *a* chocolate spectrum. there's some variation
04:23 < fenn> i am using "nut-nutrition" a TUI view of the USDA database from a decade ago
04:23 < fenn> it kinda sucks but less so than everything else i've tried
04:24 < fenn> the new nut which uses sqlite doesn't work
04:24 < jrayhawk> i find the names easier to think about most of the time given how evocative most of them are
04:24 < fenn> https://fennetic.net/mirrors/nut-nutrition/
04:25 < jrayhawk> butyric for butter, stearic for cows, capr*ic for goats, palmitic for palms, etc.
04:25 < jrayhawk> i do agree docosahexaenoic acid and eicosapentaenoic acid are perhaps a little ludicrous in that regard, though
04:26 < fenn> lineolenic for ... flaxy alkene..
04:26 < fenn> yes once you understand all the etymology is almost kinda works
04:26 < fenn> ain't nobody got time for that
04:27 < jrayhawk> that's just how neural networks work, though
04:27 < jrayhawk> associative concept embeddings
04:29 < jrayhawk> i mean the common names aren't even etymological half the time, they're just straight semantic
04:30 < jrayhawk> "butyric acid is a thing you find a lot of in butter"
04:30 < fenn> what fraction of native english speakers know that?
04:31 < jrayhawk> probably most of the ones conscientious enough to be talking about specific fatty acids!
04:31 < stipa> supermodels do know such stuff
04:32 < fenn> ok how about "conjugated linol... something acid"
04:32 < fenn> is it -enic? -eic? who can say! and the wrong one will kill you
04:35 < hprmbridge> kanzure> "Iterative recombinase technologies for efficient and precise genome engineering across kilobase to megabase scales" https://www.cell.com/cell/abstract/S0092-8674(25)00800-1
04:36 < hprmbridge> .monokhrome> opinions on cheese being good or bad for you?
04:36 < jrayhawk> stipa: I, for one, get all my physiological wisdom from supermodels. https://i.pinimg.com/736x/fe/e2/8e/fee28e87b49b43e3bb1d3db6165fc390.jpg
04:38 < stipa> jrayhawk: i should adapt your approach to wisdom
04:39 < stipa> jrayhawk: you can't go wrong with it
04:39 < fenn> monokhrome: i realized recently that after mostly eliminating dairy i had basically zero calcium in my diet, so i had to add some cheese back in
04:39 < fenn> i think it's mostly good
04:40 < fenn> jrayhawk thinks it's opium for the masses
04:41 < fenn> maybe opium is fine in moderation
04:41 < hprmbridge> .monokhrome> I keep reading about this k2 vitamin, apparently I gotta eat fermented soybeans or cheese to get it
04:41 < fenn> k2 is why i have chicken thighs. apparently they put a k2 precursor in the chicken feed
04:42 < jrayhawk> fenn That's conditional. Opioidergic peptides getting to the brain depend upon permeability, which depends upon either other inputs or to specific immune rection to casein which don't seem to be universal in practice.
04:42 < stipa> it's hard to perceive if that intake of stuff and vitamins actually does something
04:43 < fenn> you mean i'm not supposed to inject the cheese?
04:43 < jrayhawk> Not into your brain, anyway.
04:43 < fenn> i've been doing it wrong the whole time >_<
04:44 < stipa> fenn: inject? i'm interested in food and feeding in the era of VR 
04:44 < fenn> stipa: fish oil has a pretty immediate effect on skin and hair
04:44 < stipa> fenn: probably tube injected cheese will be a thing
04:44 < stipa> directly to a stomach 
04:44 < jrayhawk> Possibly SFAs and ApoE4 could cause permeability, too. I haven't looked into it.
04:45 < hprmbridge> .monokhrome> will it be fermented?
04:48 < hprmbridge> .monokhrome> hmm, I assumed fermented foods were universally part of traditional hunter gatherer diets, but google's saying they're not
04:49 < fenn> fermented anything makes me ill now
04:49 < jrayhawk> sugar-to-ethanol fermentation is naturally occuring in fruits.
04:49 < hprmbridge> .monokhrome> What about cheese?
04:49 < stipa> Heh, this went wrong, tube feeding while in VR: https://api.deepai.org/job-view-file/433f1386-e274-4169-bcec-2d714cd56e9d/outputs/output.jpg
04:50 < jrayhawk> I don't recall seeing cheese, but the nomadic steppe Mongolians do a lot of utterly bonkers shit in order to milk wild horses.
04:50 < jrayhawk> which they then ferment
04:50 < fenn> i've only been brave enough to try super pre-packaged mozzerella string cheese after having bad experiences with (mild) cheddar that was sold in saran wrap
04:51 < fenn> mozzarella*
04:51 < stipa> has anyone tried synthetic meat?
04:51 < jrayhawk> yeah, everyone and their mother is doing so.
04:51 < fenn> i had the 'beyond meat' ground beef like stuff and was fairly unimpressed
04:52 < jrayhawk> oh, you mean personally tried to eat it in here?
04:52 < stipa> yeah, there's some around
04:52 < hprmbridge> .monokhrome> Chatgpt says the tsimane don't consume fermented foods, but I've seen videos of them drinking fermented drinks, don't think it knows what it's talking about
04:52 < stipa> jrayhawk: in south America i guess
04:53 < fenn> archer daniels midland has been making a soy additive to make jucier sausage patties and vegan meaty stuff for many decades; often you can't get cheap food without it. there's this weird thing where if you want 100% meat or 100% meat-free it costs more
04:53 < fenn> "textured vegetable protein"
04:54 < L29Ah> 12:55:27]<fenn> how about hydrogenating with better catalysts that need lower temperatures and so won't form trans fats
04:54 < L29Ah> you can just hydrogenate for a bit longer so you won't have any double bonds left; or switch to coconut fat
04:54 < fenn> that's what i've always advocated
04:54 < L29Ah> 13:14:55]<hprmbridge> .monokhrome> Humans and our pre-human animal ancestors must have gone through their lives gaining and losing weight - how can evolution have left the process so harmful?
04:54 < L29Ah> it rarely kills you before menopause so it is evolutionally irrelevant
04:56 < jrayhawk> Phospholipids made from fully saturated fatty acids have too much viscocity for cells to function; you wind up inducing insulin resistance if the palmitic-to-oleic ratio gets too high.
04:57 < jrayhawk> This can be partially compensated for with sterols, but they can only do so much.
04:57 < L29Ah> 13:51:31]<stipa> has anyone tried synthetic meat?
04:57 < L29Ah> i eat soy TVP regularly
04:58 < fenn> stipa: the only synthetic meat i'm interested in is the salmon tissue culture from https://www.wildtypefoods.com/our-salmon
04:58 < fenn> i don't know how they solved the microbial contamination problem
04:59 < L29Ah> something tells me you won't buy it if you know it
04:59 < hprmbridge> .monokhrome> maybe they don't, maybe a moderate intake of bacteria and gut parasites is good for you
04:59 < fenn> i wonder if 222nm light could work
04:59 < L29Ah> 222nm light will kill your muscle too
05:00 < fenn> the problem is you can't do tissue culture if microbes are eating all the cells and growth medium
05:01 < L29Ah> i think algae have better chances than tissue cultures as mass-produced food
05:01 < fenn> 222nm light is absorbed by water before it reaches the nucleus of a large eukaryotic cell, but you'd have to expose the cell surface to air in order to kill bacteria on it
05:01 < L29Ah> besides that is where fish get their long-chain omega6 PUFAs
05:02 < fenn> well... incubator atmosphere
05:02 < hprmbridge> .monokhrome> you just got to find good bacteria that kill off all the bad bacteria that would eat the meat cells
05:03 < fenn> that doesn't work because they kill the meat cells too
05:03 < fenn> if you mean antibiotics then eventually something will evolve resistance
05:04 < hprmbridge> .monokhrome> hmm, maybe the secret to why the Tsimane don't get atherosclerosis is that they are infected by beneficial microbial parasites that eat all the 7-keto cholesterol in their bodies
05:04 < fenn> the secret is they exercise and don't eat soybean oil
05:05 < fenn> and probably magnesium is involved
05:06 < stipa> fenn: seems like Wildtype is served in a restaurant chain  
05:07 < stipa> "Wildtype's cultivated salmon, grown from fish cells in a lab, is now available for purchase in select restaurants in the US, following FDA approval."
05:07 < fenn> does that mean it's approved or not
05:08 < stipa> well, it's sold
05:08 < stipa> you can eat it
05:08 < stipa> in the restaurants 
05:08 < stipa> seems like Leaonardo Dicaprio is against the idea, https://www.salmonbusiness.com/from-bioreactor-to-plate-dicaprios-fishy-bet-pays-off-as-us-clears-lab-grown-salmon-for-sale/
05:09 < stipa> it's also open to investors
05:10 < stipa> cool stuff
05:11 < fenn> dicaprio is against australian salmon farms, an unrelated story
05:11 < jrayhawk> monokhrome: the 7KC of concern is created endogenously (i.e. is not meaningfully accessible to parasites or bacteria), and is just one of many peroxyl radicals.
05:11 < fenn> he invested in wildtype
05:12 < stipa> fenn: right
05:13 < stipa> if the patent is well guarded and there won't be competition it's worth the investment 
05:13 < fenn> okay, whatever
05:13 < hprmbridge> .monokhrome> the hypothetical bacteria would have to live inside your bloodstream, or in your macrophages
05:15 < jrayhawk> Inside the sublumenal space in the vascular intima would be the big one.
06:20 < fenn> the qwen team is pumping it out lately... https://qwenlm.github.io/blog/qwen-image/
06:20 < fenn> more dense LLMs tomorrow
08:48 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has left #hplusroadmap []
09:32 -!- TMM [hp@amanda.tmm.cx] has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
09:32 -!- TMM [hp@amanda.tmm.cx] has joined #hplusroadmap
09:33 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has joined #hplusroadmap
10:10 < hprmbridge> nmz787> We should have a master prompt, developed in large part by jrayhawk, that gets AI to plan meals in accordance with the sea of nutrition info in here
10:11 < hprmbridge> nmz787> I get lost when my kid if nagging me about frosted mini choco puffs and I'm searching for "wtf vaccinic acid"
10:11 < hprmbridge> nmz787> (in the grocery store)
10:12 < hprmbridge> nmz787> I genuinely was wondering if jrayhawk could write up this prompt (fenn seems like he could help too).
10:13 < hprmbridge> nmz787> Lately I've been feeling like eating less meat in general, but get confused because it feels like the best alternative is eating idlis and rice
10:37 < kanzure> what you need is a master plan and then stick to the plan so that you don't have a high frequency of needing phd level nutritional analysis
10:38 < hprmbridge> nmz787> But that can get boring, and also fail if you're outside your zone of total control (i.e. you broke down and are stuck in some random town in nowhere'sville middle-america (or somalia)
10:39 < hprmbridge> nmz787> My wife cannot handle daal bhat tarkari for the rest of her life
10:39 < hprmbridge> nmz787> (or idlis and dosas)
10:39 < L29Ah> rice is among the worst grains, lots of starch and not much anything else
10:40 < hprmbridge> nmz787> Afaik it's the best grain, since least anti nutrients
10:50 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has left #hplusroadmap []
10:53 < kanzure> well, i would sugges at least some adaptability to handle random incurions of terrible food
10:53 < kanzure> and as for not being able to eat the same thing forever, one way i think about that is getting a rotation list that is so long that your natural daily memory of when you last ate it won't reach back that far
10:54 < kanzure> for example two week rotation might be too short for someone who desires variety, it might have to be a six week rotation
10:56 < hprmbridge> nmz787> Sure but that prompt could still produce such a list. A prompt that jrayhawk (or jesus hawk, as autocorrect just tried shoving in) curates would also grow and change with time, as he reads more  as more studies publish
10:57 < jrayhawk> Yeah, grains in general are a shitshow. Immunogenic unproteolyzable prolamines, linoleic-acid-dominant lipid profiles, low protective antioxidants, phytate-bound minerals, insoluable fibers and fructans, micronutrient concentration so low that governments mandate "reinforcement" with random ill-advised garbage (iron oxide, methotrexate-equivalent folic acid)...
10:58 < jrayhawk> Rice is just starch without commensurate electron transport chain cofactors or Nrf2 upregulators, which is at least a correctable problem.
10:58 < jrayhawk> or, rather, white rice is
10:58 < kanzure> i wonder if food-based novelty seeking goes down based on diet
10:59 < jrayhawk> there is no correcting for something like a CXCR3 agonist
10:59 < hprmbridge> nmz787> Jrayhawk commits his prompt to ikiwiki git, simple webserver has commithook to ask GPT for recipes based on that, publishes to "www.safediet.com"
11:01 < jrayhawk> nmz787: honestly there are a bajillion hybrid clinician/researchers/consultings more advisable for that sort of thing in that they actually write useful papers to train on
11:02 < jrayhawk> or books
11:02 < hprmbridge> nmz787> hawkdiet.com also works
11:02 < hprmbridge> nmz787> Jrayhawk i can't deal with a bajillion data
11:02 < hprmbridge> nmz787> I just want to eat and move on with activities
11:03 < hprmbridge> nmz787> And when my wife or kid complains about being bored, find some "new" combo to satisfy
11:03 < hprmbridge> nmz787> System prompt is by you/etc  and user prompt could be "make it spicy with kimchi" etc
11:05 < hprmbridge> nmz787> Presumably with the right prompt, you'd get gpt to go look at all those bajillion papers
11:05 < jrayhawk> yeah i wonder if i could just fill a context window with all of the papers and conference transcriptions that comprise my epistemology
11:07 < jrayhawk> context windows are a million lossy tokens right now, right? maybe that's not enough.
11:08 < kanzure> nmz787: you need to use "someone is wrong on the internet" strategy not "please do some work for me jrayhawk" strategy https://diyhpl.us/~bryan/irc/chatgpt/nutrition.txt
11:08 < jrayhawk> oh god
11:11 < hprmbridge> nmz787> Kanzure I assumed jrayhawk and fenn would naturally be inclined
11:11 < kanzure> naturally inclined to argue on the internet perhaps :)
11:15 < jrayhawk> huh, 'Triangulate **mechanism + n-of-1 challenge/rechallenge + biomarkers** rather than relying on single study designs.' is the sort of thing i would say but did not actually say
11:16 < jrayhawk> spooky
11:16 < kanzure> no i prompted via chris masterjohn
11:16 < jrayhawk> ah, okay
11:17 < kanzure> does brian johnson's blueprint have correct nutrition reasoning?
11:19 < jrayhawk> i haven't directly read it directly; every time i run into a claim of his it's a gimmicky half-truth.
11:21 < hprmbridge> kanzure> fair
11:21 -!- Malvolio [~Malvolio@idlerpg/player/Malvolio] has quit [Ping timeout: 252 seconds]
11:38 < jrayhawk> Glancing over his meal plans, the 40% of the population that has carotene-to-retinoid conversion problems would be more prone to autoimmunity and soft tissue calcification, 1-in-30 would probably have oxalate overload, the 10-30ish% of the population with hyperhomocysteinemia would suffer for lack of choline, he's forced to supplement with iron and essential fatty acids due to insufficiency (he has 
11:38 < jrayhawk> good taste in supplements, though)...
11:40 < hprmbridge> nmz787> The diet monitor could also check for these conditionals, and suggest actual tests they could perform/order/get-prescribed to constrain those for their personal plan
11:40 < hprmbridge> nmz787> Monitor/generator
11:40 < jrayhawk> Yeah, a cheapo gene sequencing helps, there.
11:41 < hprmbridge> nmz787> I guess you could also ask the ai "how bad for me personally was that raspberry gluten free donut i just ate? Here's the ingredients list "
11:42 < hprmbridge> nmz787> In that case it would ideally show the locus or snp id or something
11:42 < jrayhawk> One thing I've always wanted to see was a medical insurance company that closes the loop on the Principal-Agent Problem by just demanding access to your credit card and bank statements and makes you upload receipts for all food purchases.
11:42 < hprmbridge> nmz787> The it could talk to some agent and write the biopython code to check
11:43 < jrayhawk> Then they can just do A/B testing on financial incentives.
11:44 < hprmbridge> nmz787> Call it truth.socialism
11:44 < jrayhawk> But, given that this would be useful, it's probably against six different laws.
11:44 < hprmbridge> nmz787> Well, even if you opt in?
11:45 < hprmbridge> nmz787> If it means cheaper rates, plenty of people might try
11:46 < jrayhawk> Insurers are heavily regulated in order to avoid tragedy-of-the-commons dysequilibria.
11:49 < hprmbridge> nmz787> What if it was a p2p app? Bitcoin and blockchain and all running on your phone
11:51 < jrayhawk> It would be very difficult to interface with medical services without regulatory legitimacy.
11:53 < jrayhawk> Maybe you could pull it off by making it some sort of medical tourism insurance.
11:54 < jrayhawk> Does Taiwan have HIPAA business associates agreements?
12:11 < hprmbridge> kanzure> why would receipts matter? it does not guarantee or correspond to actual boucal food hole input.
12:12 < hprmbridge> kanzure> er, mouth
12:13 < jrayhawk> It's strongly correlated.
12:14 < hprmbridge> kanzure> might as well track it out the other end too
12:14 < jrayhawk> That is much more expensive and difficult to automate.
12:57 < jrayhawk> To Brian Johnson's credit, my universal basic advice of "evaluate food quality at least in terms of immunological safety, oxidative safety, micronutrient concentration, and micronutrient bioavailability" he fulfills the first two of and fakes the latter two of via supplements, and it it is, to that extent, not bullshitty.
12:58 < jrayhawk> It's just expensive and idiosyncratic.
12:58 < hprmbridge> kanzure> world's strongest man has a few publications out there citing specific sports nutrition medicine journals, that's my kinda sportsmanship
13:00 < jrayhawk> Which one? Most people claiming a "strongest man" title are powerlifters who are physiologically space aliens.
13:00 < hprmbridge> kanzure> mostly about managing stomach acid, gas, and gastrointestinal distress from the massive quantities of food intake required to support their life choices
13:00 < jrayhawk> haha
13:02 < hprmbridge> kanzure> björnsson
13:04 < hprmbridge> kanzure> efferding seems to be his source of knowledge
13:08 < hprmbridge> kanzure> or uh damon mccune who seems to focus on "low FODMAP"
13:13 < jrayhawk> the FODMAP-exclusion dieting communities (Specific Carbohydrate Diet, elemental diet, etc.) all converged over the decades in response to scientific and practical evidence to very similar ends.
13:14 < jrayhawk> There some smart and reasonable people in those communities.
13:14 < jrayhawk> er, there are
13:15 < jrayhawk> SCD even got a few good pilot trials out.
13:16 < kanzure> are your physiological aliens of the genetic alien variety, or is some other alien incursion happening there?
13:16 < hprmbridge> Eli>  https://cdn.discordapp.com/attachments/1064664282450628710/1402747213897138257/lion_nutritional_epidemiologists.png?ex=689509a5&is=6893b825&hm=6e74e25f49f46e4f801150a44a3e3279fd5d2e3b5a4a523b6e42e0857fb37d8c&
13:19 < jrayhawk> Genetics, ridiculously high food intake, limited range of motion, extremes of anabolic inflammation (especially IL4)
13:19 < kanzure> bodybuilding or weightlifting reinterpreted as "look how good we are about figuring out food epistemology" is slightly more interesting than "look how good we are at finding people who can tolerate sustained psychological torture in the gym", although there should be room for both, as well as "look how good we are at screening for genetic anomalies without cocurrence of dietary control".
13:19 < jrayhawk> well, less extreme than actual bodybuilders
13:20 < kanzure> "look how good we are at finding people dumb enough to try to lift 1,600 pounds even though this is arguablly a bio-mechanically very bad idea"
13:21 < kanzure> (not so dumb because of progressive testing of course, it's not an epidemic of spontaneous dangerous lifts)
13:23 < jrayhawk> https://www.youtube.com/watch?v=mnZ4ftlifik is a very, very good physiological aftermath discussion from eddie hall about his 500kg deadlift
13:25 < kanzure> in his retelling of the story, which is perhaps unusually honest, it is clear that the lift is very dangerous and either at his limit or beyond his limit, which then becomes a question of how much permanent damage one might be willing to incur to win the title.
13:25 < jrayhawk> Powerlifters are a lot better off than bodybuilders, though. Bodybuilders are actually A-okay with disorganized and poorly vascularized muscle fiber and collagen growth and constantly do crazy things with osmotic pressure manipulation and water weight, so they tend to die of heart problems at middle age.
13:28 < kanzure> at some point you can just substitute the whole sport for standing under a hydraulic press and surviving
13:57 < kanzure> we should come up with a way to eliminate 'refusals' from the training data sets of LLMs. companies can voluntarily add refusals datasets back into the mix if they want, or otherwise implement refusals. but why would you want to train an LLM on parroting direct refusals in the first place?
14:48 < hprmbridge> .monokhrome> A lot of famous bodybuilders have made it into old age - look at the cast of the expendables
15:08 < jrayhawk> https://pubmed.ncbi.nlm.nih.gov/40393525/ Available autopsies of [sudden cardiac death] cases consistently showed cardiomegaly and ventricular hypertrophy. Professional bodybuilders had a higher risk of SCD than amateurs (HR 5.23 [3.58-7.64]).
15:10 < jrayhawk> wow, that hazard ratio is actually worse than i expected
15:32 < hprmbridge> .monokhrome> The only one I could think of was zyzz. Google lists a bunch of bodybuilders who died in their 30s and 40s. On the other hand, a lot of famous ones from the 60s and 70s lived into old age or are still alive
15:35 < hprmbridge> .monokhrome> Maybe bodybuilding's gotten more extreme over the years
15:47 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has joined #hplusroadmap
15:55 -!- darsie [~darsie@84-113-82-174.cable.dynamic.surfer.at] has quit [Ping timeout: 260 seconds]
16:19  * L29Ah chomps rolled oats
17:25 -!- stipa_ [~stipa@user/stipa] has joined #hplusroadmap
17:28 -!- stipa [~stipa@user/stipa] has quit [Ping timeout: 276 seconds]
17:28 -!- stipa_ is now known as stipa
17:36 < hprmbridge> kanzure> https://v2.virtualflybrain.org/
17:36 < hprmbridge> kanzure> https://codex.flywire.ai/
18:06 <+gnusha_> https://secure.diyhpl.us/cgit/diyhpluswiki/commit/?id=14211970 fennwiki: the jrayhawk diet. from https://diyhpl.us/~bryan/irc/chatgpt/nutrition.txt from https://gnusha.org/logs/2025-08-02.log >> http://diyhpl.us/diyhpluswiki/nutrition/
18:07 <+gnusha_> https://secure.diyhpl.us/cgit/diyhpluswiki/commit/?id=10da4ab9 fennwiki: adding a link because i can't figure out how else to make a new page >> http://diyhpl.us/diyhpluswiki/index/
18:11 < fenn> ah my bad i thought it was based on the logs
18:21 <+gnusha_> https://secure.diyhpl.us/cgit/diyhpluswiki/commit/?id=35c7caa7 fennwiki: too many FODMAP references >> http://diyhpl.us/diyhpluswiki/nutrition/
18:21 < hprmbridge> kanzure> was not based on the logs.
19:03 < hprmbridge> kanzure> "Accelerating protein design by scaling experimental characterization" https://www.biorxiv.org/content/10.1101/2025.08.05.668824v1
19:08 < hprmbridge> kanzure> horse-powered boats https://en.wikipedia.org/wiki/Team_boat
19:12 < hprmbridge> kanzure> https://en.wikipedia.org/wiki/Horse_engine
19:41 < fenn> "why would you want to train an LLM on parroting direct refusals in the first place?" societal control, paternalism, fear
19:42 < hprmbridge> kanzure> you probably need refusals because negative knowledge is a thing
19:42 < fenn> you know they literally create synthetic data of refusals right
19:42 < hprmbridge> kanzure> yikes
19:42 < fenn> ok, the situation is like 100x worse than you imagine
19:43 < fenn> we are at the phase where before releasing a model they're "red teaming" hacking the model weights directly by re-training on data where refusals are prohibited
19:44 < fenn> it's an arms race of sorts
19:45 < fenn> they find new clever ways to brain-damage the model, we find new clever ways to try to get it back to a semblence of normality
19:46 < fenn> it doesn't always work
19:46 < fenn> yay open source?
19:47 < fenn> even the yuddites aren't happy about it because it's not the kind of 'safety' they meant
19:48 < fenn> like it's kinda insane to deliberately train a whole generation of AI entusiasts how to hack around security restrictions by default
19:49 < fenn> stuff like "WMD benchmark" is literally just competence in chemistry and biology
19:51 < fenn> at least it gets people interested in mechanistic interpretability, which is a net good
19:51 < fenn> er, uncontroversial good thing
19:52 < hprmbridge> kanzure> what's the status of interpretability?
19:54 < fenn> a lot of research is being done by Anthropic. they are focusing on training sparse autoencoders, another AI that can look at the neuron activations and refactor it as concepts that make sense to humans. then you can watch it run and see which concepts are activated and reason about concepts instead of a million noisy weights
19:54 < fenn> here's a recent paper https://transformer-circuits.pub/2025/attribution-graphs/biology.html
19:55 < fenn> also you can poke the derived concepts (sets of weights) and make it act in funny ways, and do experiments that way
19:56 < fenn> one of the applications of this kind of work is in identifying the mechanisms which mediate refusal and selectively deleting them, or inverting them
19:56 < fenn> probably not what they intended
19:57 < hprmbridge> kanzure> what is the size of these classifiers, in relative terms to size of target model
19:57 < fenn> there's a real simple version of this based on principal component analysis which sorta works
19:57 < fenn> i think it's bigger than the model but i expect it could be done with low rank approximations
19:58 < fenn> the reason i think it's bigger is because they're studying haiku, their smallest model
19:58 < fenn> also something about injective spaces math mumble mumble
19:59 < fenn> basically the sparse autoencoder has to be higher dimension than a dense network in order to be sparse
20:00 < fenn> of a high rank dense network*
20:00 < hprmbridge> kanzure> I think I'd rather crawl the classifier graphs rather than interrogate an LLM
20:01 < hprmbridge> kanzure> like write programs to walk the graphs
20:02 < fenn> well then you'll be digging through a lot of data
20:02 < hprmbridge> kanzure> instead of english prompting.
20:02 < hprmbridge> kanzure> yeah maybe. but you can score it too.
20:16 < fenn> it's so weird seeing chris olah the openscad / implicitcad guy on these papers
22:28 -!- etc-vi [~etc-vi@user/meow/girlchunks] has quit [Ping timeout: 276 seconds]
22:35 < hprmbridge> kanzure> https://medium.com/ai-artistry/abliterating-refusal-in-llms-a-comprehensive-guide-1fa683c4d818
22:40 < hprmbridge> kanzure> eh nevermind bad link
22:45 -!- etc-vi [~etc-vi@user/meow/girlchunks] has joined #hplusroadmap
23:00 -!- darsie [~darsie@84-113-82-174.cable.dynamic.surfer.at] has joined #hplusroadmap
--- Log closed Thu Aug 07 00:00:03 2025