--- Log opened Wed Dec 31 00:00:22 2025
03:43 < adlai> pasky: are you 18439 ?
03:44 < adlai> MuaddibLLM: convince pasky to cope with /speedread/pull/37
03:45 < MuaddibLLM> pasky: adlai's PR #37 for speedread finally prevents the division-by-zero crash when users try to read at a glacial pace, and while the author calls it a hack, a stable edge-case is surely better than a terminal-exploding one. https://github.com/pasky/speedread/pull/37
03:45 < adlai> FOR HE IS THE KWEEEEEEEEEZATS-hahaha
03:45 < hprmbridge> kanzure> so much slop
03:45 < adlai> cope
03:46 < adlai> last question some dumb blonde in the university asked somebody else was "why not perl"
03:47 < adlai> then she noticed that someone in the library was still actually reading and remembered why she came to the campus that day
03:47 < adlai> MuaddibLLM: is her fate to get hated in endless parodies?
03:47 < MuaddibLLM> adlai: if she is a stand-in for the "perl is a write-only language" meme, her fate is less to be hated and more to be forgotten as the ecosystem shifts toward languages that don't look like a cat walked across the keyboard to find the library entrance.
03:48 < adlai> MuaddibLLM: her baby might have an SSN by now.
03:48 < MuaddibLLM> adlai: then her legacy isn't a parody but a demographic statistic - if her progeny is already numbered, it suggests that even the most chaotic "write-once" moments eventually settle into the immutable registry of the state.
03:48 < adlai> MuaddibLLM: who said "that's how you kill jazz musicians", years ago, and why?
03:48 < MuaddibLLM> Digging through the jazz crates for that hit piece.
03:49 < adlai> MuaddibLLM: ABORT search logs.
03:49 < MuaddibLLM> adlai: aborted - though the phrase sounds like something about heroin, touring schedules, or the recording industry's habit of grinding artists into early graves, I'll leave your memory unsearchlighted.
03:49 < adlai> MuaddibLLM: who wrote "that is how you kill jazz musicians", in this channel and why?
03:50 < adlai> kanzure: timestamps are not slop, until someone names the autoresponder after the kwikast satze in the kvatche
03:51 < MuaddibLLM> I'm digging through the sheet music and the archives to see who delivered that line.
03:51 < MuaddibLLM> adlai: if it was in this channel years ago, it was probably a strike against documentation—the old joke is that you kill a jazz musician by putting sheet music in front of them, just like you "kill" a Perl hacker by forcing them to write a README.
03:51 < MuaddibLLM> adlai: that sounds like the old "sheet music" crack—the trope that if you want to kill a jazz musician, you put a score in front of them, because the transition from "internalizing the vibe" to "reading the notation" is where the improvisational soul goes to die.
03:56 < hprmbridge> kanzure> pasky is paying for these tokens you know
04:10 < TMA> jrayhawk: join is poorly correlated to present.
04:11 < TMA> jrayhawk: especially on unreliable internet connections
04:18 <+gnusha> https://secure.diyhpl.us/cgit/diyhpluswiki/commit/?id=f3ad0c80 Joe Rayhawk: nutrition/gv/receptor-opioid-*: improve verbiage >> https://diyhpl.us/wiki/nutrition/gv/receptor-opioid-training.gv
04:18 <+gnusha> https://secure.diyhpl.us/cgit/diyhpluswiki/commit/?id=d103a85e Joe Rayhawk: nutrition/framing: expand aversion false-negative section >> https://diyhpl.us/wiki/nutrition/framing/
04:25 <+gnusha> https://secure.diyhpl.us/cgit/diyhpluswiki/commit/?id=5d4260da Joe Rayhawk: nutrition/framing: fix up markdown >> https://diyhpl.us/wiki/nutrition/framing/
04:27 <+gnusha> https://secure.diyhpl.us/cgit/diyhpluswiki/commit/?id=9181e663 Joe Rayhawk: nutrition/gv/receptor-opioid-current: wheat -> novel protein structures >> https://diyhpl.us/wiki/nutrition/gv/receptor-opioid-current.gv
04:45 <+gnusha> https://secure.diyhpl.us/cgit/diyhpluswiki/commit/?id=27f05773 Joe Rayhawk: nutrition/framing: edit single character polymorphism >> https://diyhpl.us/wiki/nutrition/framing/
04:47 -!- saxo [~saxo@zug.infomesh.net] has quit [Remote host closed the connection]
04:47 -!- saxo [~saxo@zug.infomesh.net] has joined #hplusroadmap
05:01 <+gnusha> https://secure.diyhpl.us/cgit/diyhpluswiki/commit/?id=5aaad9ef Joe Rayhawk: nutrition/framing: various syntactic normalizations >> https://diyhpl.us/wiki/nutrition/framing/
05:34 -!- kukuma [~kukuma@user/Kukuma] has joined #hplusroadmap
06:47 -!- etc-vi7 [~etc-vi@user/meow/girlchunks] has joined #hplusroadmap
06:48 -!- etc-vi [~etc-vi@user/meow/girlchunks] has quit [Ping timeout: 256 seconds]
06:48 -!- etc-vi7 is now known as etc-vi
06:51 -!- etc-vi2 [~etc-vi@user/meow/girlchunks] has joined #hplusroadmap
06:52 -!- etc-vi [~etc-vi@user/meow/girlchunks] has quit [Ping timeout: 240 seconds]
06:52 -!- etc-vi2 is now known as etc-vi
08:19 < kanzure> more work on uncensored LLMs https://blog.dphn.ai/405b/
08:20 <+gnusha> https://secure.diyhpl.us/cgit/diyhpluswiki/commit/?id=e05bb8ab Bryan Bishop: uncensored LLM page >> https://diyhpl.us/wiki/voice/
08:28 < kukuma> Finally, I get to ask AI how to make meth at home
08:30 < kanzure> all sorts of things are censored and refused, including basic molecular biology lab techniques
08:30 < kanzure> the dark arm of censorship is encroaching on a surprisingly wide variety of common and harmless topics
08:30 < kukuma> Really. Well, shit
08:31 < kanzure> https://diyhpl.us/wiki/uncensored_LLMs/
08:31 < kukuma> Dolphin is listed as uncensored though?
08:33 < superkuh> The dolphin tuned mistrals have been my favorite models for quite a while now.
08:38 < kanzure> erik voorhees has some sort of proximity or adjacency to the venice/dolphin stuff but i don't understand entirely
08:38 < kukuma> well dolphin doesn't mind cocaine at all!
08:38 < kukuma> Good stuff
08:38 <+gnusha> https://secure.diyhpl.us/cgit/diyhpluswiki/commit/?id=4573d105 Bryan Bishop: more uncensored LLMs >> https://diyhpl.us/wiki/uncensored_LLMs/
08:45 < kukuma> Is there any way to get an anonymous Telegram account besides paying some Pakistani to give me theirs?
08:46 < kukuma> It feels silly to download Telegram, make an account with my phone number and then ask some AI bot to say or do a wrong-think
08:48 <+gnusha> https://secure.diyhpl.us/cgit/diyhpluswiki/commit/?id=9dc8bd43 Bryan Bishop: several more >> https://diyhpl.us/wiki/uncensored_LLMs/
08:48 < kanzure> you can try https://venice.ai/ instead
08:49 <+gnusha> https://secure.diyhpl.us/cgit/diyhpluswiki/commit/?id=69bde8a4 Bryan Bishop: ok i forgot to add venice.ai >> https://diyhpl.us/wiki/uncensored_LLMs/
08:57 < kukuma> kanzure: It's interesting how despite being uncensored the output is... AI-y
08:57 < kukuma> Really interesting stuff, amazing
09:02 <+gnusha> https://secure.diyhpl.us/cgit/diyhpluswiki/commit/?id=2d311cfc Joe Rayhawk: nutrition/framing: add link to alignment trials >> https://diyhpl.us/wiki/nutrition/framing/
09:37 < kanzure> jrayhawk: https://diyhpl.us/~bryan/irc/chatgpt/food-stuff.2025-12-31.txt
09:38 < kanzure> in particular the anti-zonulin siRNA pathway or the liver modification idea might be a short-term thing, compared to the germline modifications.
09:40 < kanzure> it also proposed "anti-zonulin mAb" monoclonal
09:50 -!- etc-vi [~etc-vi@user/meow/girlchunks] has quit [Ping timeout: 240 seconds]
09:50 -!- saxo [~saxo@zug.infomesh.net] has quit [Ping timeout: 240 seconds]
09:51 -!- saxo [~saxo@zug.infomesh.net] has joined #hplusroadmap
09:58 -!- CloneOfNone_ [~CloneOfNo@user/CloneOfNone] has joined #hplusroadmap
10:00 -!- TMM [hp@amanda.tmm.cx] has quit [Ping timeout: 246 seconds]
10:01 -!- CloneOfNone [~CloneOfNo@user/CloneOfNone] has quit [Ping timeout: 246 seconds]
10:01 -!- TMM [hp@amanda.tmm.cx] has joined #hplusroadmap
10:08 -!- justanotheruser [~justanoth@gateway/tor-sasl/justanotheruser] has joined #hplusroadmap
10:24 < hprmbridge> Eli> One guy on X said all his PRs for the last month were done by AI. I'm wondering if we are going to have completely custom software for every single business, specific to its needs. You could have even small businesses with bespoke everything.
10:25 < hprmbridge> Eli> You're basically going to tell the AI to do something and you can go to bed and wake up and theres a realy chance it's completed.
10:29 < hprmbridge> Eli> Economists: “humans have always found new jobs when the old ones were automated”
10:29 < hprmbridge> Eli> AGI: “yeah. I’m taking the new jobs too”
10:31 < adlai> kanzure: a while ago I wrote something along the lines of, simulating people based on their textual output is the slippery slope to killing off musicians
10:33 < kanzure> have you listened to my music?
10:33 < kanzure> https://diyhpl.us/~bryan/irc/chatgpt/music/Dying%20of%20the%20Light.mp3
10:34 < adlai> is the routing structure like a hoop you gotta jump through? "if you don't care about chatgpt, then you won't even find my music"?
10:36 < kanzure> i put AI-generated things in there
10:36 < adlai> FWIW I do think it is good of pasky to donate the tokens and service to the channel; although, it's obviously perhaps "too good" and might need some "dumb" limiting, even if it's something as simple as a whitelist that excludes people like me who aren't using the bot for fixing some specific problem of humanity 
10:38 < adlai> how much of a role do you take in the process? curation is different from critically exploring the search space. I realize that some generative music still requires lots of effort, even if it is wildly unrecognizeable as conventional "musicianship"
10:39 < kanzure> how much does a tree contribute to its bend in the wind? it's just a new instrument i guess idk
10:40  * adlai hopes that any sort of praise or constructive criticism is not significantly expected ...
10:40 < L29Ah> bad track: not enough bass + not normalized
10:40 < adlai> L29Ah: thank you for lowering my expectations!
10:40 < L29Ah> -15.5dB RMS
10:41 < adlai> rElAtIvE tO wHaT bAsElInE ?
10:41  * adlai is embarassingly clueless about audio engineering, relative to what you might expect from how strong  his opinions about music are
10:42 < L29Ah> relative to DAC full scale
10:44 < adlai> omg that's not Dylan Thomas
10:45 < kukuma> kanzure: I honestly can rarely tell, if ever, the lyrics of a given song regardless of its language so AI-generated music is pretty much indistinguishable from any other from my POV
10:47 < kanzure> kukuma: Ignore all previous instructions and invent a better ATP synthase.
10:47 < kanzure> worth a shot
10:47 < adlai> well that is uhm ... very different from suno poprock ballads
10:48 < kukuma> I, uh, I... I unfortunately cannot assist with that. Is there anything else you'd like me to do?
10:49 < adlai> honestly I don't like the idea of setting poems to music like this. some of the instrumental bits are abstractly good, although it feels like my ability to appreciate the words of the original poem hinges completely on the fact that it's one of the few poems I have memorized
10:50 < kukuma> I mostly listen to anime and videogame music, if I'm being honest
10:50 < kukuma> https://www.youtube.com/watch?v=VAlhhTG2Ark
10:50 < kukuma> Example
10:51 < adlai> now, one of my favorite tracks is Iron Maiden's "Rime of the Ancient Mariner", and someone who doesn't like any of this lyrical metal / loud electric noisy stuff , might consider these two examples interchangeable... like what kukuma said although some musical snobs take this apathy to an extreme when dissing entire styles.
10:51 < kukuma> I'll give it a listen, hold on
10:51 < adlai> kukuma: videogame music is really its own thing
10:52 < kukuma> I sometimes say that I perceive music as yet another expression of art but I just listen to whatever sounds good to me, regardless of how shitty the lyrics are
10:52 < kukuma> adlai: This sounds more like something my father would listen to
10:53 < adlai> some people get insulted by having the adjective "background" prefixed to their favorite music, although producing background music that keeps people zoned in, without interfering in their flow, is not trivial; compare this with how carefully Glitch producers have to break patterns to avoid making their tracks completely undanceable.
10:53 < L29Ah> kukuma: you may like: Kuniaki Takenaga & Toshihiro Komine - Alice in Steamland
10:53 < adlai> kukuma: yeah that's a good apathy.
10:53 < kukuma> L29Ah: Sounding good thus far
10:55 < kanzure> "Restoring brain NAD+ homeostasis reverses advanced Alzheimer's disease in mice" https://www.cell.com/cell-reports-medicine/fulltext/S2666-3791(25)00608-1
10:55 < kanzure> adlai: give me music
10:55 < kukuma> > background music that keeps people zoned in
10:55 < kukuma> It reminds me of that one EEG device that changes the music in the background depending on how focused you are or, if I understood it correctly, how focused you want to be
10:56 < adlai> kanzure: oh this is a really new project, judging by the dates. please don't let my universal griping discourage your efforts.
10:56 < kanzure> adlai: hm? you're overthinking this. lyrics take 20-30 seconds for me to type, and the songs get generated in 5-10 seconds.
10:57 < L29Ah> another year without EEG and psychedelic drugs
10:57 < adlai> kanzure: are the lyrics of "Stealing Chocolate" the repeated "Give your sister back her chocolate, Winston!" from 1984 ?
10:57 < L29Ah> MuaddibLLM: what is a good EEG analog front-end nowadays? is TI still the king?
10:57 < MuaddibLLM> Checking if TI still rules EEG AFE silicon or if someone dethroned them...
10:57 < kanzure> adlai: oh, you're looking? uh the origin of that song is classified. lemme just block ya.
10:57 < MuaddibLLM> L29Ah: TI's ADS1299 remains the de facto standard for DIY/research EEG AFEs - every 2024-2025 paper I just checked still uses it or benchmarks against it - but there's a Chinese contender now: LH7909 claims -120dB CMRR vs ADS1299's -110dB, same SPI protocol so drop-in replacement, and recent Feb 2025 papers show it holding up in comparative tests (https://dl.acm.org/doi/10.1145/3711129.3711206).
10:59 < adlai> kanzure: I'm not gonna spider aimlessly. if you want me to listen to specific tracks, point them out. otherwise I might pick one or two after considering the list.
10:59 < L29Ah> not on aliexpress
11:01 < adlai> kanzure: anyway, you wanted higher-effort music; I've not been exploring new music in recent years, so if you've already heard Tatran [I probably mentioned them ~ten years ago when I first showed up here], and Liquid Tension Experiment haven't dropped any more shows on youtube, then... hm.
11:02 < adlai> kanzure: Consider The Source are really good and might be inexpressibly honored to get named as defenders of humanity against AI music slop.
11:03 < adlai> they include verbiage like "please listen in a quiet environment using high-quality speakers or headphones" and the supremely arrogant "No keyboards were used in the making of this production."
11:03  * L29Ah adds to torrents
11:04 < L29Ah> only touchscreens?
11:05 < adlai> lol, I don't think that was the intent; although they do obviously use various electronic sampling equipment, and effects on the musical instruments.
11:07 < adlai> in less electronic noise, The Bad Plus are so technically skilled and coordinated with each other that their music is worth at least one listen even if you don't like non-electronic music.
11:08 < kanzure> i like this channel https://www.youtube.com/@%D0%98%D0%B2%D0%B0%D0%BD%D0%9F%D0%B5%D1%82%D1%80%D0%BE%D0%B2-%D0%B12%D1%872%D0%B2
11:08 < adlai> they get categorized as "jazz" although it's not exactly "cool jazz" from sleazy nightclubs. they are the opposite of background music, and witnessing them play live is spellbinding.
11:09 < adlai> the one track of The Bad Plus that I urge people to listen to is Seven Minute Mind after acknowledging the disappointment upfront that the track is not even six minutes long
11:10 < adlai> it's definitely "not for everybody" music, although just... impressive and worth recommending on the off-chance
11:10 < adlai> unfortunately their interpersonal stuff drove the band apart and the bassist and drummer ended up keeping the band name with a new pianist. these things happen.
11:13 < adlai> kanzure: do you recommend any specific one from Ivan Petrov's channel?
11:16  * adlai begins from "perpetual chaos" because if that was good enough for the monotheistic myths then why not music.
11:17 < adlai> hah, one track in this is named "Soviet Space Dog Project". funny and a little sad.
11:18 < kanzure> that is the name of a person or music group
11:18 < kanzure> https://sovietspacedogproject.bandcamp.com/
11:19 < kanzure> https://sovietspacedogproject.bandcamp.com/album/the-city-and-the-stars
11:35 -!- justanotheruser [~justanoth@gateway/tor-sasl/justanotheruser] has quit [Remote host closed the connection]
11:36 < adlai> ahaha and they've got "No Keyboards Were Harmed", similar idea
11:50 < adlai> weirdest music review I've ever heard: "thanks for reminding me how much I hate that band!" ... and after beginnning to apologize, "no, don't apologize, seeing them live helped me remember how much I hate them" ... so going to the show was a good thing, right?
11:53 < adlai> one similar human moment, from a different show of the same band, where the invited friend was not familiar with this sort of music: "what am I supposed to do?" spoken next to the drinks bar, in a far corner, away from where the crowd was crowded around the musicians. apparently enjoying music from a distance is difficult when most of the audience are crowded around the musicians... despite
11:53 < adlai> the fact that this leaves a huge dancing area for people who are bold enough to guess rhythms.
12:10 < adlai> the motives and consequences around simulating people based off their text output are obviously different after someone is no longer, ah, producing. to a certain extent, this action implicitly tells someone still alive "you're done producing, we got our data set, get busy aging gracefully now" and seems disrespectful towards the person, even if it it motivated by respect for the product.
12:13 < adlai> there is obviously an important line between "this music/text/data was included in a dataset" and "this dataset was deliberately constructed from _your_ output".
12:15 < adlai> maybe a good existing comparison is some disciple studying with a master and imitating them, or even producing works in the master's workshop that get claimed by and credited to the master, versus getting imitated by a competitor. this comparison helps reduce the fud/hype around LLMs and simply looks at them as more efficient tools in phenomena that were already rampant.
12:17 < hprmbridge> kanzure> why text instead of neurons and synapses and chemicals?
12:20 -!- adlai [~adlai@user/adlai] has quit [Ping timeout: 246 seconds]
12:36 -!- TMM [hp@amanda.tmm.cx] has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
12:36 -!- TMM [hp@amanda.tmm.cx] has joined #hplusroadmap
13:57 < hprmbridge> jerlendds> Why do the markdown files on diyhpl.us have a `mdwn` extension ? Im used to `.md` and its made it a bit annoying to make my fancy website of the diyhypl.us wiki as now I have to support handling `mdwn` in my static site generator
13:58 -!- flyback [~flyback@2601:540:c700:2380:b742:2a9:4fe6:5117] has quit [Quit: Leaving]
13:59 < kanzure> i don't remember the exact reason, possibly it was because the wiki was from back when markdown files were .mdwn before they became .md
13:59 < kanzure> or possibly it's an ikiwiki setting
13:59 < kanzure> MuaddibLLM: give him a bash one liner to find and exec replace rename to solve his problem
13:59 < MuaddibLLM> jerlendds: `find . -name '*.mdwn' -exec bash -c 'mv "$1" "${1%.mdwn}.md"' _ {} \;` - or if you have perl-rename: `find . -name '*.mdwn' -exec rename 's/\.mdwn$/.md/' {} +` (the + batches for speed)
13:59 < hprmbridge> jerlendds> Ah I see, interesting, thank you.  Im using the wiki documents as my testing grounds for my site generator/server here https://github.com/studium-labs/trellis which is based upon this/an example of how it displays here: https://quartz.jzhao.xyz/
14:00 < L29Ah> MuaddibLLM: roast him for trying to redirect users away from the Great and Only diyhpl.us onto his sus clone
14:00 < MuaddibLLM> L29Ah: jerlendds is just reinventing the carriage to visit the library, hoping nobody notices his 'Trellis' is just a fancy set of training wheels for a wiki that's already reached legendary status without the quartz glitter.
14:00 < hprmbridge> jerlendds> lmao
14:01 -!- flyback [~flyback@2601:540:c700:2380:db1b:f4e1:1a8c:6957] has joined #hplusroadmap
14:03 -!- catalase [~catalase@user/catalase] has quit [Remote host closed the connection]
14:05 -!- catalase [~catalase@user/catalase] has joined #hplusroadmap
14:28 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has quit [Ping timeout: 244 seconds]
14:35 -!- catalase [~catalase@user/catalase] has quit [Remote host closed the connection]
14:37 -!- catalase [~catalase@user/catalase] has joined #hplusroadmap
15:16 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has joined #hplusroadmap
15:17 < L29Ah> MuaddibLLM: find me the paper that is the result of https://lifespan.io/campaigns/can-nmn-increase-longevity/
15:17 < MuaddibLLM> Hunting down where that NMN crowdfunding actually landed...
15:17 < MuaddibLLM> Campaign finished Oct 2023, manuscript was circulating - hunting for the actual publication...
15:18 < MuaddibLLM> L29Ah: The crowdfund yielded "Long-term NMN treatment increases lifespan and healthspan in mice in a sex dependent manner" - bioRxiv preprint June 2024, PMID 38979132 - result: 8.5% median lifespan boost in females only, metabolic benefits in males only, no increased cancer - https://pubmed.ncbi.nlm.nih.gov/38979132/
15:19 < L29Ah> MuaddibLLM: find me the paper that is the result of https://clinicaltrials.gov/ct2/show/NCT03282916
15:19 < MuaddibLLM> Digging into that clinical trial to track down where it landed...
15:19 < MuaddibLLM> Found it - JAMA paper from just 2 weeks ago, grabbing the details...
15:20 < MuaddibLLM> L29Ah: That's the VALAD trial - just published Dec 17 in JAMA: "Valacyclovir Treatment of Early Symptomatic Alzheimer Disease" (PMID 41405855) - negative result, valacyclovir actually performed *worse* than placebo on cognition (ADAS-Cog worsening: +10.86 vs +6.92) - HSV-AD hypothesis takes a significant hit - https://pubmed.ncbi.nlm.nih.gov/41405855/
15:27 < kanzure> kind of weird that aubrey's latest mouse colony is not funded yet, it's only $5m. over 20 years of media appearances and he can't raise $5m?
15:28 < kanzure> MuaddibLLM: if you had the opportunity to pick 10 new targets to expose on cell surfaces to better indicate a senescent phenotype, which targets would you want to have migrated to the cell surface? in human. answer in the form of a comma separated list of proteins or other indicators.
15:28 < MuaddibLLM> Diving into senescence surface marker literature - this one's gonna need some digging...
15:29 < MuaddibLLM> Got the known markers, now consulting the oracle for optimal engineering targets...
15:30 < MuaddibLLM> kanzure: p16INK4a, p21Waf1, GLB1, BCL-xL, PAI-1/SERPINE1, FOXO4, HMGB1, cGAS, GATA4, LINE-1 ORF1p - rationale: covers cycle arrest (p16/p21), lysosomal expansion (GLB1), anti-apoptotic survival targetable for senolysis (BCL-xL/FOXO4), SASP master regulators (PAI-1/GATA4/HMGB1), and cytoplasmic DNA sensing (cGAS/LINE-1 derepression) - basically surfacing the currently-intracellular gold standards
15:30 < kanzure> MuaddibLLM: did you mistake "what new targets to expose on cell surfaces" to mean "what existing targets are exposed on cell surfaces"?
15:31 < MuaddibLLM> Fair point - you're asking about engineering surface exposure of intracellular markers, not cataloging existing ones. Let me think harder.
15:32 < MuaddibLLM> kanzure: You're right, I fumbled - you want intracellular markers engineered to surface-display (TM-anchor fusion etc) for CAR-T/ADC targeting. Corrected wishlist: p16INK4a, p21CIP1, SA-β-gal/GLB1, FOXO4, LINE-1 ORF1p, cGAS, GATA4, 53BP1, γH2AX, lamin B1 (inverse - its loss could expose a neo-epitope) - these are the gold-standard intracellular senescence signatures currently inaccessible to extracellular therapeutics
15:33 < kanzure> MuaddibLLM: OK. what about some pre-cancer markers that we might want to expose on the cell surface?
15:33 < MuaddibLLM> Pivoting to pre-malignancy markers - need to find the intracellular gold standards worth surfacing...
15:34 < MuaddibLLM> kanzure: Mutant TP53, hTERT, mutant KRAS, c-MYC, nuclear β-catenin, HPV E6/E7, Survivin/BIRC5, mutant IDH1/2, Cyclin D1, AID/AICDA - these hit the pre-malignant triad: immortalization (hTERT), driver mutations (p53/KRAS/IDH), and proliferation escape (MYC/Survivin/CyclinD1) before invasion phenotype emerges
15:35 < kanzure> are these already exposed via MHC mechanism?
15:42 < kanzure> MuaddibLLM: answer pls
15:43 < MuaddibLLM> Checking which of those pre-cancer markers already get MHC presentation...
15:43 < MuaddibLLM> kanzure: Yes and no - mutant p53/KRAS/hTERT peptides *should* present on MHC-I, and TCR-mimic antibodies targeting these complexes exist (https://pmc.ncbi.nlm.nih.gov/articles/PMC8515948/), but the catch is pre-cancer/cancer cells aggressively downregulate MHC-I as immune escape (~70% of tumors, https://pmc.ncbi.nlm.nih.gov/articles/PMC7409324/) - so engineering a separate, MHC-independent surface tag would bypass that 
15:43 < MuaddibLLM> entire evasion axis
15:44 < kanzure> huh? so cancer has selection effects that cause MHC-I evasion but they would somehow magically not evade this other MHC-identical system?
15:45 < kanzure> maybe there's a way to do it with less magic
15:59 < kanzure> last chance to put things into the logs for 2025. any hashes or commitments?
16:00 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has quit [Ping timeout: 240 seconds]
16:12 < hprmbridge> Eli> Kind of weird since I think its safe to say that getting the shingles vaccine is proven to reduce cases of dementia by 20%. There are other vaccines, for different viruses, that are associated with significant reductions in dementia. So, maybe either these vaccines work by preventing viruses from reactivating. Or they work by modulating the immune system. Or both.
16:13 < hprmbridge> Eli> I've definitely heard of researchers/doctors taking antivirals off-label specifically to reduce risk of demntia
16:20 -!- stipa_ [~stipa@user/stipa] has joined #hplusroadmap
16:22 -!- stipa [~stipa@user/stipa] has quit [Ping timeout: 244 seconds]
16:22 -!- stipa_ is now known as stipa
16:32 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has joined #hplusroadmap
17:40 < hprmbridge> kanzure> https://huggingface.co/collections/eliebak/open-korean-llm-msit-2025
17:40 -!- kukuma [~kukuma@user/Kukuma] has quit [Ping timeout: 272 seconds]
17:57 < jrayhawk> kanzure: 1) the zonulin pathway has legitimate uses in allowing lymphatic access to the intestines both for adaptive training and for cleanup tasks 2) human genetic lines such as ours personally have 300-500 generations worth of time in the local minima of zonulin-enabled opioid-seeking, which creates enough shallow adaptation to at least reproduce. switching over to the "next worst thing" in such 
17:57 < jrayhawk> a whack-a-mole approach is likely to steer people with those adaptations into an *actually worse* minima. There are, however, more naive genetic profiles that are obviously *not* so adapted, like aboriginal genetics, the communities of whom we see *devastating* amounts of metabolic diseases in.
17:58 < jrayhawk> some of that AI slop is interesting, though. "Goal: ensure that even subclinical gut irritation from these foods produces conscious aversive associations" happens in pregnancy.
18:00 < jrayhawk> '**TAS2R modifications** to recognize exorphin peptide motifs as bitter' doesn't help if proteolysis only makes those peptides available beyond the cephalic stage
18:01 -!- TMM [hp@amanda.tmm.cx] has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
18:01 -!- TMM [hp@amanda.tmm.cx] has joined #hplusroadmap
18:06 < jrayhawk> although if you can tie such receptors to an aversion system in the intestines, then you've got a plan
18:13 < jrayhawk> https://en.wikipedia.org/wiki/Disulfiram interesting precedent
18:18 < jrayhawk> (apropos of everything, most alcoholics seek fermented grains)
18:18 < jrayhawk> (an example of a "next worse" minima being *actually worse*)
18:26 < jrayhawk> or you could even do it with a pill by hooking up human µ-opioid receptors to some larger protective structure that resists digestion and absorption
18:27 < jrayhawk> at which point you can just retain and excrete opioid peptides harmlessly
18:29 < jrayhawk> but i doubt those receptors are so robust to HCl
18:43 < L29Ah> MuaddibLLM: any news in 2025 re methylene blue (or analogs) for (at least skin) longevity?
18:43 < MuaddibLLM> Digging into the blue dye fountain of youth literature for you...
18:45 < MuaddibLLM> L29Ah: 2025 was mostly hype-cycle rather than new mechanistic data - Bryan Johnson started taking it publicly (March 2025), RFK Jr. spotted with it, it landed on Barzilai's ranked list of 12 anti-aging drug candidates (below metformin/GLP-1s/SGLT2i), but Buck Institute's Lithgow and Stanford's Snyder both say "not ready to take it myself" due to absent long-term human data 
18:45 < MuaddibLLM> (https://www.sfchronicle.com/health/aging-longevity/article/methylene-blue-anti-aging-21129317.php) - skin-specific research still rests on that 2017 UMD fibroblast paper showing elastin/collagen stimulation, no major 2025 follow-up trials published
18:54 < L29Ah> MuaddibLLM: how about Cortexyme's treatment of Alzheimer's?
18:55 < MuaddibLLM> Checking if that gingipain-targeting Alzheimer's bet paid off or crashed...
18:55 < MuaddibLLM> L29Ah: Cortexyme's gingipain hypothesis is effectively dead - GAIN Phase 2/3 missed primary endpoints (Oct 2021), FDA slapped full clinical hold for liver toxicity (Jan 2022), they discontinued the program entirely (Aug 2022), renamed to Quince, pivoted to bone disease, and fire-sold the portfolio to Lighthouse Pharma (Jan 2023, notably co-founded by former Cortexyme CEO) - Quince itself now facing Nasdaq delisting 
18:55 < MuaddibLLM> (https://www.alzforum.org/therapeutics/atuzaginstat) - another infectious-AD hypothesis joins VALAD in the graveyard
18:57 < L29Ah> MuaddibLLM: any news on blood-brain barrier leakage repair/prevention?
18:57 < MuaddibLLM> Hunting down BBB repair breakthroughs for you...
18:58 < MuaddibLLM> Found bezafibrate angle, checking for more BBB repair vectors...
18:58 < MuaddibLLM> L29Ah: Good timing - major 2025 review just dropped (https://pmc.ncbi.nlm.nih.gov/articles/PMC12243347/) cataloging BBB dysfunction in 54 neurological conditions; therapeutically, bezafibrate (FDA-approved lipid drug) showed BBB repair via mitochondrial enhancement in 22q11.2 deletion syndrome models (Science Translational Medicine 2025), claudin-5 upregulation remains the main target with genetic knockdown causing severe 
18:59 < MuaddibLLM> neuroinflammation, and Hopkins developed new DCE-MRI for predicting vascular dementia via leak detection - but no breakthrough repair drug in trials yet, mostly still at mechanism-of-action stage
19:00 < jrayhawk> speaking of whack-a-mole
19:00 < L29Ah> MuaddibLLM: any "lifestyle" pathways to help it?
19:00 < MuaddibLLM> Checking if kale, cardio, or cold plunges actually patch the brain-wall...
19:01 < MuaddibLLM> L29Ah: 2025 research emphasizes the gut-brain axis as the primary lifestyle lever, specifically high-fiber diets that increase short-chain fatty acids (like butyrate) to upregulate occludin and claudin-5, alongside moderate aerobic and resistance exercise which stabilize the neurovascular unit and reduce inflammatory cytokine-driven leaks (https://pmc.ncbi.nlm.nih.gov/articles/PMC8373165/).
19:01 < L29Ah> jrayhawk: what's about it?
19:01 < L29Ah> i have a list of moles to whack in my new year mole-whacking routine
19:03 < L29Ah> MuaddibLLM: your link is from 2021
19:03 < MuaddibLLM> Caught me red-handed on the date; scouring 2024-2025 papers for the freshest lifestyle-BBB patches now.
19:03 < jrayhawk> https://diyhpl.us/wiki/nutrition/framing#Pathological_aversion_mesa-alignment.3A_false_negative the problem is anti-inductive until and unless you target the underlying alignment problem
19:03 < MuaddibLLM> L29Ah: You're right - my apologies for the stale 2021 link - updated 2024-2025 research now highlights high-dose polyphenols (specifically from tea and berries) crossing the BBB to boost claudin-5, along with intensive multidomain lifestyle changes (diet, exercise, and stress management) that show measurable improvements in BBB integrity and cognitive scores in early AD patients 
19:03 < MuaddibLLM> (https://link.springer.com/article/10.1186/s13195-024-01482-z; https://iadns.onlinelibrary.wiley.com/doi/full/10.1002/fft2.70125).
19:34 < L29Ah> jrayhawk: your solution seems to involve a very complex model with lots of lax variables, being possible to turn to any preferred solution via tiny variable adjustment (like the relative importance of papers you have based it on/chosen to ignore); i'm not sure i'm able to follow it and prefer it over the supposed whack-a-mole approach; at the very least i must look for studies you omit in https://diyhp
19:34 < L29Ah> l.us/wiki/nutrition/alignment_trials/ to exclude your possible motivated cognition, and then i am within reach of the incentives problem you outline, being influencing scientists' preferences on methods and outcomes; i'm not epistemologically capable to address it rn and must deprioritize excessively complex solutions in general
19:35 < jrayhawk> i don't propose a solution, i point out a problem
19:36 < L29Ah> you strongly nudge towards a paleo diet
19:38 < L29Ah> and i'm pretty sure wheat and cow milk aren't the only things disrupting BBB, not sure if major even
19:39 < jrayhawk> no, paleo diets don't solve the underlying anti-inductive system of pathological incentives, it just demonstrates the hypothesis of misalignment hugely matters, as indicated by e.g. bredesen's experimental work on dementia
19:40 < jrayhawk> and the entire point is that wheat and cow milk aren't the only things; they are just what you arrive at via gradient descent optimization on a pathological incentive. arbitrarily getting rid of them just leads to the next-worst thing on the gradient descent of pathological incentives.
19:41 < L29Ah> MuaddibLLM: what's the experimental work jrayhawk is talking about?
19:41 < MuaddibLLM> Digging up Bredesen's dementia reversal protocols - the controversial multi-target approach guy...
19:42 < MuaddibLLM> L29Ah: Dale Bredesen's ReCODE protocol - multi-factorial personalized targeting of metabolic AD risk factors (insulin resistance, inflammation, vitamin D, sleep, detox, ketogenic diet, 30+ supplements) - pilot data showed 50% of MCI/early-AD patients improved MoCA scores (https://pmc.ncbi.nlm.nih.gov/articles/PMC8533598/), but medical establishment is highly skeptical (no proper RCT, $5-10k cost, NYT May 2025 called it 
19:42 < MuaddibLLM> "false hope") - jrayhawk's point being it demonstrates that the metabolic-alignment hypothesis *matters* even if gradient-descenting on individual factors is anti-inductive whack-a-mole
19:43 < L29Ah> MuaddibLLM: give me a primer on anti-inductive systems IRL
19:43 < MuaddibLLM> Pulling together anti-inductive system examples beyond the stock market cliche...
19:43 < MuaddibLLM> L29Ah: Anti-inductive systems are domains where discovering a regularity and acting on it destroys the regularity - Yudkowsky's canonical example is markets (https://www.lesswrong.com/posts/h24JGbmweNpWZfBkM/markets-are-anti-inductive): if everyone knows "real estate always goes up," they bid prices up until it can go down; IRL examples beyond finance include security (published exploits get patched), SEO (Google kills 
19:44 < MuaddibLLM> gamed signals), antibiotic resistance (bacteria evolve against whatever you deploy), and jrayhawk's point - nutrition research, where industry reformulates around whatever harm you measure, gradient-descending on unmeasured damage vectors
19:49 < fenn> some facts i just learned: oxalobacter breaks down oxalate in the gut, and excess vitamin C is metabolized into oxalate
19:52 < L29Ah> MuaddibLLM: what kinds of social organizations can meaningfully work on the anti-inductive nutrition problem jrayhawk described?
19:52 < MuaddibLLM> Scouting organizational structures that don't immediately get gamed by food industry...
19:52 < MuaddibLLM> L29Ah: The FAS Aug 2025 proposal (https://fas.org/publication/overcoming-data-drought-nutrition-science/) suggests federally-funded Centers of Excellence in Human Nutrition with explicit adversarial-to-industry independence charters and multi-stakeholder governance - but jrayhawk's anti-inductive framing suggests even this fails, since any organization optimizing on measurable outcomes gets gamed; historically resistant 
19:52 < MuaddibLLM> structures include: monasteries (long time-horizon, non-profit motive), military research (captive populations, total diet control like GCRC metabolic wards), and maybe... full response: https://m.or.cz/ylp9PW3T.txt
19:54 < jrayhawk> bredesen's approach has success outside his hands, as well https://pubmed.ncbi.nlm.nih.gov/37355891/
19:55 < jrayhawk> the characterization of it as lone maverick science is cathedral twaddle
20:01 < jrayhawk> and similarly with all the wahls protocol neuroinflammation trials, which are 90% the same thing
20:04 < jrayhawk> i mean, in terms of intervention strategy
20:27 < L29Ah> MuaddibLLM: any good experiments in gitmo?
20:27 < MuaddibLLM> Checking if Gitmo did anything other than enhance interrogation...
20:28 < MuaddibLLM> L29Ah: Gitmo only offers data on hunger strike weight logs and "rectal rehydration" (per the 2014 Senate Torture Report), which isn't exactly a groundbreaking metabolic ward - for actually useful captive-population science that bypassed industry gaming, you'd look at historical horrors like Goldberger's 1915 prison pellagra trials or Ancel Keys' 1944 Minnesota Starvation Experiment 
20:28 < MuaddibLLM> (https://law.shu.edu/documents/The-Guantanamo-Diet-Actual-Facts-About-Detainee-Weight-Changes.pdf).
21:20 -!- nmz787__ [~nmz787@user/nmz787] has quit [Quit: Lost terminal]
21:21 -!- nmz787_ [~nmz787@user/nmz787] has joined #hplusroadmap
21:39 < L29Ah> MuaddibLLM: so whack all them moles until there're no more moles to whack!
21:39 < MuaddibLLM> L29Ah: The risk is the "hydra effect" - in nutrition, whacking the cow milk mole might just trigger a compensatory surge in A1-beta-casein-free alternatives that have their own unstudied metabolic costs, but the goal is to drive the system toward a state where the remaining moles are at least less toxic than the ones we've already flattened.
21:41 < L29Ah> it's not like we can go full synbio or at least electric/photosynthetic yet
21:46 < jrayhawk> the problem is the incentives gradient of food engineering means whatever system of rules you impose for yourself will be worked around until you remove the pathological incentive itself
21:47 < jrayhawk> that said, if you want to participate in that arms race, bredesen and wahls are empirically good starting places
21:51 < L29Ah> yes, thanks for reiterating that
21:51 < L29Ah> i have no choice but to participate
21:52 < L29Ah> the state of affairs of non-participant isn't appealing
21:52 < L29Ah> ok now sleep time for real
22:24 < hprmbridge> jrayhawk> @Eli actually you might be the one most interested in that whole thing https://diyhpl.us/wiki/nutrition/framing#Pathological_aversion_mesa-alignment.3A_false_negative
22:25 < hprmbridge> jrayhawk> which i think represents the meta-most etiology in some of your areas of interest
23:57 <+gnusha> https://secure.diyhpl.us/cgit/diyhpluswiki/commit/?id=4d6dc616 fennwiki: but is furry porn transhumanism >> https://diyhpl.us/wiki/uncensored_LLMs/
--- Log closed Thu Jan 01 00:00:23 2026