--- Log opened Sun Aug 31 00:00:26 2025
00:04 -!- darsie [~darsie@84-113-82-174.cable.dynamic.surfer.at] has joined #hplusroadmap
00:26 < fenn> https://cdn.arstechnica.net/wp-content/uploads/2025/08/starship.jpg
00:26 < fenn> red color is from some metallic test tiles that oxidized
00:33 < fenn> docl the ganges river is full of bacteriophages from all the corpses dumped in it...
00:37 < fenn> snapping turtles tear them apart within a few minutes and presumably a lot of bacteria is released during this process
00:42 < fenn> From the day we arrive on the planet, And blinking, step into the sun, There's more to see than can ever be seen, More to do than can ever be done, There's far too much to take in here, More to find than can ever be found
00:42 < fenn> It's the ciiiiircle.. the circle of liiiiife
00:52 < fenn> 'enzymes from other phages that “chew up” biofilm' very cool
00:53 < fenn> you'd think a phage would want to keep the farm running
00:55 < fenn> today i'm making yogurt from general biotics 
00:55 < fenn> ahem
00:55 < fenn> from general biotics' "equilibrium" probiotic blend of 100 relatively unknown bacteria
00:56 < fenn> it's annoyingly expensive
00:56 < fenn> but i have a few sample pills
00:57 < fenn> smells like "friendship cake" sourdough starter
01:01 -!- stipa_ [~stipa@user/stipa] has joined #hplusroadmap
01:02 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has quit [Ping timeout: 258 seconds]
01:03 -!- MuaddibLLM [muaddib@pasky.or.cz] has quit [Ping timeout: 258 seconds]
01:03 -!- fleeky [~fleeky@prsm.space] has quit [Ping timeout: 258 seconds]
01:04 -!- geneh2 [~cam@pool-71-191-180-23.washdc.fios.verizon.net] has quit [Ping timeout: 258 seconds]
01:04 -!- superkuh [~superkuh@user/superkuh] has quit [Ping timeout: 258 seconds]
01:04 -!- stipa [~stipa@user/stipa] has quit [Ping timeout: 258 seconds]
01:04 -!- stipa_ is now known as stipa
01:05 -!- MuaddibLLM [muaddib@pasky.or.cz] has joined #hplusroadmap
01:05 -!- fleeky [~fleeky@prsm.space] has joined #hplusroadmap
01:05 -!- geneh2 [~cam@pool-71-191-180-23.washdc.fios.verizon.net] has joined #hplusroadmap
01:19 -!- superkuh [~superkuh@user/superkuh] has joined #hplusroadmap
01:25 < hprmbridge> kanzure> "Defining the cell and molecular origins of the primate ovarian reserve" https://www.nature.com/articles/s41467-025-62702-0
01:31 < hprmbridge> kanzure> "Generating brain-wide connectome using synthetic axonal morphologies" https://www.nature.com/articles/s41467-025-62030-3
04:30 < fenn> 96GB VRAM 400GB/s for $1500 ish https://www.hardware-corner.net/huawei-atlas-300i-duo-96gb-llm-20250830/
04:31 < fenn> it's also small
04:31 < fenn> and low power
04:34 < fenn> lol the "typical application scenario" is some dystopian surveillance state
04:35 < fenn> "Smart City 2.0 to identify illegal roadside stalls, monitor food safety, and implement water governance, meeting the requirements of government departments for intelligent city governance and providing important support for service decision-making."
04:36 < fenn> only vague compute performance numbers (200 INT8 TOPS)
04:37 < fenn> good enough for LLM but i'd want to know before spending money
04:42 -!- darsie [~darsie@84-113-82-174.cable.dynamic.surfer.at] has quit [Quit: Avoid fossil fuels and animal products. Have no/fewer children. Protest, elect sane politicians. Invest ecologically.]
04:42 < fenn> sorry 280 TOPS INT8
04:43 -!- darsie [~darsie@84-113-82-174.cable.dynamic.surfer.at] has joined #hplusroadmap
04:52 < fenn> Half-precision (FP16): 140 TFLOPS
04:52 < fenn> now it's $4k
04:57 < fenn> prices are all over the place on alibaba, maybe it's still <$1.5k
04:57 -!- Malvolio [~Malvolio@idlerpg/player/Malvolio] has joined #hplusroadmap
05:21 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has joined #hplusroadmap
06:03 < fenn> wow. ebay has banned capsule fillers. good job, land of the free
06:04 < fenn> and they had to pay $59 million, for ... uh, doing nothing illegal?
06:04 < fenn> because *other* people were doing illegal things
06:09 -!- TMM [hp@amanda.tmm.cx] has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
06:09 -!- TMM [hp@amanda.tmm.cx] has joined #hplusroadmap
06:09 < L29Ah> selling capsule fillers is illegal?
06:09 < L29Ah> https://www.amazon.com/capsule-filling-machine/s?k=capsule+filling+machine
06:20 < fenn> unclear what you're saying
06:21 < fenn> https://fedequipblog.fedequip.com/dea-regulations-for-tablet-press-and-capsule-filler-sales/
06:21 < fenn> pretty crazy for a flimsy piece of plastic with some holes in it
06:23 < fenn> clearly the criminals will just give up now
06:46 < pasky> hm any thoughts on the best uncensored model to add into muaddib? my default thinking went to deepseek r3.1 (the code basically works, just routing remaining to be done), but not 100% excited about sending data to china...
06:49 < pasky> maybe hermes4, but another account eh and they support only legacy openai chat completions api so i'd pull in openrouter and then i'd pull in litellm for openrouter .. it's all so tiresome
06:56 < hprmbridge> kanzure> I really like Kimi K2
06:56 < L29Ah> i would be more upset about sending data to USA if i were you, after all China has no leverages against you unlike USA that can pull you out of europe to have fun with you at its court
06:57 < L29Ah> i can't call deepseek uncensored also
06:59 < hprmbridge> kanzure> groq runs kimi if you want, which is not china
07:04 -!- s0ph1a [sid246387@id-246387.helmsley.irccloud.com] has quit [Server closed connection]
07:04 -!- s0ph1a [sid246387@id-246387.helmsley.irccloud.com] has joined #hplusroadmap
07:27 < fenn> there are lots of providers for deepseek-r1 on openrouter
07:27 < hprmbridge> kanzure> true, and you can pin the provider.
07:28 < fenn> it turns out to be important :(
07:28 < hprmbridge> kanzure> anthropic admitted to random quality degradation the other day
07:28 < fenn> "random"?
07:29 < fenn> we added special degradation crystals to your coffee to see if you'd notice?
07:30 < fenn> or, we cheaped out on inference during peak load because of global strain on chip fabs and datacenter construction rates
07:30 < hprmbridge> kanzure> https://status.anthropic.com/incidents/h26lykctfnsz
07:30 < hprmbridge> kanzure> they got caught or it was a genuine bug
07:30 < fenn> that reads like a bug report
07:30 < hprmbridge> kanzure> https://x.com/TheAhmadOsman/status/1961870729609175088
07:31 < fenn> there are a lot of subtle reasons LLMs can degrade in intelligence
07:31 < pasky> kimi k2 could be actually kinda cool, thanks for that nudge! okay, building openaichatcompletions api and openrouter interface it is
07:32 < fenn> "DMCA takedowns of repos that have to do with Claude Code"?
07:32 < fenn> that sounds pretty nasty
07:37 < hprmbridge> kanzure> probably some of the reverse engineered claude code things.
07:38 < hprmbridge> kanzure> pasky: pin the openrouter kimi provider to groq. it's super fast. cerebras also offers some fast models
07:43 < pasky> thanks!
08:00 -!- darsie [~darsie@84-113-82-174.cable.dynamic.surfer.at] has quit [Quit: Avoid fossil fuels and animal products. Have no/fewer children. Protest, elect sane politicians. Invest ecologically.]
08:01 -!- darsie [~darsie@84-113-82-174.cable.dynamic.surfer.at] has joined #hplusroadmap
08:26 < hprmbridge> kanzure> Tongyi lab, moonshot, zhipu, Shanghai ai lab, IA-CAS, Deepseek, stepflow (stepfun), Baichuan, Huawei Pangu, bytedance, NLP of Tsinghua u (MINICPM-V), Hunyuan, kunlun tech, FMB, OpenGVlab (InternLM)
08:42 < pasky> ...is a list of...?
08:42 < L29Ah> MuaddibLLM: answer
08:43 < MuaddibLLM> pasky: a who’s-who list of Chinese AI labs/providers and their model families - handy for sourcing non‑US, sometimes looser models via OpenRouter, Groq, Cerebras, etc.
09:21 <+gnusha> https://secure.diyhpl.us/cgit/diyhpluswiki/commit/?id=3862b2f4 Bryan Bishop: add DNA nanoscopy references. >> http://diyhpl.us/diyhpluswiki/mapseq/
09:41 -!- justanotheruser [~justanoth@gateway/tor-sasl/justanotheruser] has quit [Ping timeout: 272 seconds]
09:46 <+gnusha> https://secure.diyhpl.us/cgit/diyhpluswiki/commit/?id=88323ed1 Bryan Bishop: expansion microscopy >> http://diyhpl.us/diyhpluswiki/in-situ-sequencing/
09:51 -!- justanotheruser [~justanoth@gateway/tor-sasl/justanotheruser] has joined #hplusroadmap
09:54 < kanzure> what was this from https://diyhpl.us/wiki/images/TKTL1_corticogenesis.png
10:07 <+gnusha> https://secure.diyhpl.us/cgit/diyhpluswiki/commit/?id=f80352bf Bryan Bishop: rearrange chairs on deck of titanic >> http://diyhpl.us/diyhpluswiki/protein_engineering/
10:28 <+gnusha> https://secure.diyhpl.us/cgit/diyhpluswiki/commit/?id=979c2430 Bryan Bishop: various updates, also some interesting protein pfam families >> http://diyhpl.us/diyhpluswiki/reservoir_computing/
10:30 < kanzure> oh yeah, the DNA ticker tape memory stuff... that should be added somewhere.
10:36 < pasky> 2025-08-31 18:58:00,970 - irssi_llmagent.main - INFO - [TEST MODE] Generated response for #hplusroadmap: kanzure: it’s from Science, 9 Sep 2022 - the Perspective “Scaling brain neurogenesis across evolution” illustrating the TKTL1 Lys-to-Arg change in modern humans, see https://www.science.org/doi/10.1126/science.ade4388 (alt PDF via Ovid: 
10:36 < pasky> https://www.ovid.com/journals/scie/pdf/10.1126/science.ade4388~scaling-brain-neurogenesis-across-evolution), summarizing the primary article https://www.science.org/doi/10.1126/science.abl6422.
10:39 < pasky> i think i'll enable the proactive interjections, it looks fairly sane, lmk if it's noisy (also if you tell muaddib to shut up he'll hopefully respect it as long as in the 30-message context window)
10:50 <+gnusha> https://secure.diyhpl.us/cgit/diyhpluswiki/commit/?id=2144b408 Bryan Bishop: more optogenetics >> http://diyhpl.us/diyhpluswiki/optogenetics/
11:14 -!- stipa_ [~stipa@user/stipa] has joined #hplusroadmap
11:17 < hprmbridge> kanzure> "Rapid learning of neural circuitry from holographic ensemble stimulation enabled by model-based compressed sensing" https://www.biorxiv.org/content/10.1101/2022.09.14.507926v3.abstract
11:17 -!- stipa [~stipa@user/stipa] has quit [Ping timeout: 256 seconds]
11:17 -!- stipa_ is now known as stipa
11:21 < hprmbridge> kanzure> another compressive sensing article https://www.cell.com/patterns/fulltext/S2666-3899(23)00224-6
11:28 -!- justanotheruser [~justanoth@gateway/tor-sasl/justanotheruser] has quit [Ping timeout: 272 seconds]
12:48 -!- EnabrinTain_ [sid11525@id-11525.helmsley.irccloud.com] has quit [Server closed connection]
12:48 -!- EnabrinTain_ [sid11525@id-11525.helmsley.irccloud.com] has joined #hplusroadmap
13:27 -!- gl00ten [~gl00ten@2001:8a0:7ee5:7800:46d9:f5c:17a2:432] has quit [Ping timeout: 248 seconds]
13:34 < pasky> kanzure: groq seems to be ratelimited *a lot* :(
13:35 < pasky> also seems to have a bad rep based on a quick google?
13:39 -!- pasky` [pasky@2a01:9f40:a000:ff00::1] has joined #hplusroadmap
13:39 < pasky`> jc, my shell's internet died
13:39 < pasky`> anyway, i'll try going with fireworks as primary, together+groq as fallbacks
13:40 -!- pasky [~pasky@nikam.ms.mff.cuni.cz] has quit [Ping timeout: 255 seconds]
13:42 -!- pasky` is now known as pasky
13:45 < hprmbridge> kanzure> "Survival of transplanted allogeneic beta cells with no immunosuppression" https://www.nejm.org/doi/10.1056/NEJMoa2503822
13:46 < hprmbridge> kanzure> compare to:
13:46 < hprmbridge> kanzure> "Transplantation of chemically induced pluripotent stem-cell-derived islets under abdominal anterior rectus sheath in a type 1 diabetes patient" https://www.cell.com/cell/fulltext/S0092-8674(24)01022-5
13:46 < hprmbridge> kanzure> which we discussed several months ago IIRC.
13:52 -!- pasky_ [~pasky@nikam.ms.mff.cuni.cz] has joined #hplusroadmap
13:57 -!- pasky is now known as pasky`
13:57 -!- pasky_ is now known as pasky
14:03 < jrayhawk> beta cells need to be in close proximity to alpha cells in order to release insulin at high enough concentration to shut down alpha cell glucagon production
14:04 -!- gl00ten [~gl00ten@87.196.73.1] has joined #hplusroadmap
14:04 < jrayhawk> T1D is, in practice, as much a disease of glucagon excess as it is of insulin deficiency
14:10 -!- gl00ten2 [~gl00ten@nat-vpn-fm.net.ul.pt] has joined #hplusroadmap
14:12 -!- gl00ten [~gl00ten@87.196.73.1] has quit [Ping timeout: 256 seconds]
15:05 < hprmbridge> kanzure> https://github.com/torinmb/mediapipe-touchdesigner
15:08 -!- darsie [~darsie@84-113-82-174.cable.dynamic.surfer.at] has quit [Ping timeout: 252 seconds]
16:06 < hprmbridge> kanzure> "Anti–USAG-1 therapy for tooth regeneration through enhanced BMP signaling" https://pmc.ncbi.nlm.nih.gov/articles/PMC7880588/
16:12 -!- gl00ten2 [~gl00ten@nat-vpn-fm.net.ul.pt] has quit [Ping timeout: 256 seconds]
16:12 -!- gl00ten [~gl00ten@87.196.73.1] has joined #hplusroadmap
16:13 < hprmbridge> kanzure> seems unlikely to scale to primate tooth regeneration but whatever. BMP/Wnt strikes again.
16:17 -!- gl00ten [~gl00ten@87.196.73.1] has quit [Ping timeout: 258 seconds]
16:19 < pasky> MuaddibLLM: !u  provide basic agrobacterium culture instructions
16:19 < MuaddibLLM> Basic Agrobacterium culture cheat-sheet:; ; 1. Medium: Use YEP (10 g/L yeast extract, 10 g/L peptone, 5 g/L NaCl, pH 7.0) or LB if you’re lazy; add 15-20 g/L agar for plates.; 2. Antibiotics: Whatever plasmid you’re carrying probably wants 25 µg/mL Rif + 50 µg/mL Kan (check specific resistance cassettes).; 3. Starter: Inoculate single colony into 2–5 mL liquid YEP, 28 °C, 250 rpm, overnight → mushy, 
16:19 < MuaddibLLM> straw-colored turbidity.; 4. Scale: 1:100 into larger flask, re-incubate to OD600 ≈ 0.6–0.8 (late log) if you need competent or density-matched inoculum.; 5. For plant infiltration: Pellet and gently resuspend in 10 mM MgCl₂ + 10 µM acetosyringone to induce Vir genes.; 6. Storage: Add 15% sterile glycerol to 0.4–0.6 OD culture → aliquot to ‑80 °C; streak from frozen when lazy.; (Full verified protocol: Wood et
16:20 < pasky> it's terrible at writing on IRC and it still won't tell you how to make napalm
16:20 < pasky> but i guess it's a bit better and it's fun to read kimi k2 in other contexts
16:25 < hprmbridge> kanzure> it's a fun model sir
16:32 -!- gl00ten [~gl00ten@bl5-239-125.dsl.telepac.pt] has joined #hplusroadmap
16:40 -!- Gooberpatrol66 [~Gooberpat@user/gooberpatrol66] has quit [Quit: Konversation terminated!]
16:45 -!- Gooberpatrol66 [~Gooberpat@user/gooberpatrol66] has joined #hplusroadmap
17:04 < fenn> hmph: https://ikiwiki.info/todo/redirect_automatically_after_rename/
17:06 < fenn> actually, i'd rather not have loads of duplicate filenames lying around, since that would be confusing
17:06 < fenn> why were they ever named foo-bar in the first place? seems like that wouldn't have happened if links didn't work right
17:07 < kanzure> it originated from my personal filenaming habit for my local filesystem
17:07 < kanzure> i don't name files gene_editing.txt
17:08 < fenn> maybe it's easier to edit ikiwiki to first try foo_bar and then foo-bar
17:08 < fenn> then urls don't change, commit history is preserved, etc
17:12 <+gnusha> https://secure.diyhpl.us/cgit/diyhpluswiki/commit/?id=a183a6a4 Bryan Bishop: DNA ticker tape memory >> http://diyhpl.us/diyhpluswiki/optogenetics/
17:17 < kanzure> "Recording gene expression order in DNA by CRISPR addition of retron barcodes" https://pmc.ncbi.nlm.nih.gov/articles/PMC9357182/
17:17 < kanzure> "retro-cascoder" is a new one to me: "Here we report a molecular system for making time-ordered recordings of transcriptional events into living genomes. We do this via engineered RNA barcodes, based on prokaryotic retrons, which are reverse-transcribed into DNA and integrated into the genome using the CRISPR-Cas system. The unidirectional integration of barcodes by CRISPR integrases enables ...
17:18 < kanzure> ...reconstruction of transcriptional event timing based on a physical record via simple, logical rules rather than relying on pre-trained classifiers or post-hoc inferential methods."
17:20 < kanzure> it's really dumb that so much of the progress in lineage tracing, projectome mapping, connectome mapping, DNA molecular ticker tape memory, etc, are all driven by trying braindead simple recombinations of existing gene editing technologies and it's whoever writes the grant fast enough to try "hey what about that thing that was done last year but with this new Cas9 variant!" who gets to write ...
17:20 < kanzure> ...the next thing. we should be able to be far more systematic and intentional about progress in this area.
17:21 < kanzure> do we even have a whole organism cell lineage history atlas yet? throughout the whole of development, sliced by developmental biology pause point?
17:24 < kanzure> and all the new gene editing things are mostly downstream of being clever about fusion proteins of existing gene editors or methyltransferases
17:24 < kanzure> ... most of which you can probably predict by combinatorially generating a list of new fusion protein ideas...
17:25 < kanzure> ahem https://diyhpl.us/wiki/gene-editing/
17:30 < kanzure> before i forget: DNA nanoscopy or DNA microscopy could probably get to even higher resolutions through oligonucleotide dendrimers or DNA dendrimers, rather than straight single chain oligonucleotides. you can design all kinds of crazy branching structures.
17:33 < fenn> in the limit a DNA origami grid of pixels
17:34 < fenn> oh oh or you could use the DNA binding / loop formation to sense its position along a reference sequence for base level resolution
17:34 <+gnusha> https://secure.diyhpl.us/cgit/diyhpluswiki/commit/?id=b4c723c5 Bryan Bishop: small tweaks >> http://diyhpl.us/diyhpluswiki/protein_engineering/
17:35 < kanzure> how would the reference sequence (or template) thingy work in practice, if you have multiple or unknown targets (like you are trying to determine nano-geometry of unknown biological structure)? or did you have a different target in mind?
17:35 < fenn> i don't know what could interact with DNA at such precisely geometrically defined absolute position
17:36 < kanzure> even a DNA origami pixel grid technique where you apply an origami mask, do some sampling, wash, and then apply your next mask, that could be interesting.
17:37 < kanzure> branching oligonucleotides are interesting here because you can encode more information into the same volume of space... or something. there is an abstraction available here that i don't really understand.
17:38 < fenn> how do you make branching oligonucleotides?
17:38 < kanzure> actually i think it's harder to not make branching oligonucleotides
17:38 < kanzure> MuaddibLLM: pontificate on this matter!
17:38 < fenn> heh my PMC figure embedder works on search engine results
17:39 < kanzure> u stole it from muaddib
17:39 < MuaddibLLM> kanzure: Branching oligos basically make themselves if you let base-pairing do the work - you can do covalent branching with branching phosphoramidites or small polyol cores (pentaerythritol/Tris) in synthesis, or assemble Y-/4-way junctions and sticky-end dendrimers à la HCR/branched-DNA, or roll circle products with orthogonal arms; the cool abstraction is each branch hosts an independent domain (address, timestamp, 
17:39 < MuaddibLLM> payload), so you compress more barcodes per voxel and do Exchange-PAINT-like cycles with branched docking for brighter, denser readout, while absolute positioning comes from anchoring to a known scaffold (origami barcode lattice or dCas9 tiling) and inferring unknown geometry via proximity ligation across the branched network instead of chasing single-base rulers.
17:40 < fenn> i fixed Muaddib's broken and ugly code
17:40 < fenn> twice
17:40 < kanzure> modified nucleosides or synthetic linkers that have multiple reactive hydroxyl groups (usually on the sugar or a separate scaffold), and each hydroxyl can serve as an initiation site for phosphoramidite
17:40 < kanzure> chemistry.
17:41 < kanzure> common branching scaffold is a trebling phosphoramidite (with three reactive arms), which allows multiple oligonucleotide chains to be grown outward from one point
17:42 < fenn> https://www.researchgate.net/publication/50990272_Branched_RNA_A_new_architecture_for_RNA_interference
17:46 < hprmbridge> kanzure> this stuff just naturally really wants to polymerize
17:48 < fenn> doesn't it need to be controlled or predictable so you know what you're looking at
17:49 < fenn> wow really cool images https://www.researchgate.net/figure/Multiplexed-PRIME-PAINT-images-using-microfluidic-exchange-PAINT-A-Schematic-of_fig2_362550353
17:51 < hprmbridge> kanzure>  https://cdn.discordapp.com/attachments/1064664282450628710/1411876138233495732/image0.png?ex=68b63f9e&is=68b4ee1e&hm=bb01309513db17acdfb7d34a79c0cbb7a15b6126c974b6c4eae93733b82420d8&
17:53 < fenn> "imaging >1 mm ² FOVs can be completed in as little as an hour per target" the highest resolution image is 15 microns in diameter so the resolution would be in the tens of nanometers
17:53 < fenn> that'll scan your connectome!
17:55 < fenn> i still haven't really understood the -PAINT concept but apparently it can do 3d images
17:55 < fenn> which would mean thicker slices, less disruption from microtomy
17:56 < MuaddibLLM> fenn: DNA-PAINT is transient binding of short fluorescent imagers to complementary docking strands on your targets so each site blinks repeatedly for nm-scale localization, 3D comes from z-encoding optics like astigmatism, biplane or spinning-disk confocal, and multiplexing uses Exchange-PAINT by swapping orthogonal imagers - good explainers and 3D demos here: https://pmc.ncbi.nlm.nih.gov/articles/PMC7612686/ and 
17:56 < MuaddibLLM> https://pmc.ncbi.nlm.nih.gov/articles/PMC5727263/.
17:57 < fenn> ok the "transient" is really short apparently. so you get a single isolated gaussian blob to localize
17:59 < hprmbridge> kanzure> also someone should consider XNAs for DNA nanoscopy
18:04 < hprmbridge> kanzure> "spatial sequencing"
18:09 < hprmbridge> kanzure> "expansion sequencing" oh
18:10 < hprmbridge> kanzure> which is, unfortunately, unrelated to the xpandomer expansion sequencing chemistry from stratos and roche
18:11 < fenn> "Alon et al. combined expansion microscopy with long-read in situ RNA sequencing," <- that one?
18:12 < hprmbridge> kanzure> ya
18:12 < fenn> .t https://pmc.ncbi.nlm.nih.gov/articles/PMC7900882/
18:12 < saxo> Expansion Sequencing: Spatially Precise In Situ Transcriptomics in Intact Biological Systems - PMC
18:12 < hprmbridge> kanzure> expansion sequencing is not on https://diyhpl.us/wiki/expansion_microscopy/
18:13 < hprmbridge> kanzure> but to my credit mentioned the idea on https://diyhpl.us/wiki/brain_uploading/ didn't know it was called "expansion sequencing"
18:13 < fenn> words
18:14 < fenn> you have some operators, you can apply them in various permutations
18:22 < fenn> so DNA-PAINT doesn't use sequencing, just base pair hybridization. Exchange-PAINT gets more color channels by washing out the DNA-dyes and using new dyes with different binding targets. there's probably a bunch of schemes that use sequencing to get more color channels. "MERFISH" is one i guess
18:24 < hprmbridge> kanzure> here's a MERFISH technique with branched DNA https://www.nature.com/articles/s41598-019-43943-8
18:25 < fenn> no MERFISH is "spatial transcriptomics" meaning it's only imaging RNA that's already present
18:26 < fenn> MuaddibLLM: is there a thing like DNA-PAINT where you are imaging proteins using gaussian centroid of fluorescent sequencing as the imagers?
18:26 < MuaddibLLM> Hunting for protein PAINT methods that use sequencing as the blinking mechanism...
18:27 < MuaddibLLM> r.jina.ai HTTP 500, retrying in a bit...
18:27 < MuaddibLLM> r.jina.ai HTTP 500, retrying in a bit...
18:29 < MuaddibLLM> Found peptide-PAINT paper but site's being grumpy - checking other sources for protein PAINT variants...
18:29 < MuaddibLLM> r.jina.ai HTTP 500, retrying in a bit...
18:29 < hprmbridge> kanzure> pasky: these error messages aren't that helpful to the audience.
18:29 < MuaddibLLM> r.jina.ai HTTP 500, retrying in a bit...
18:31 < fenn> peptide-paint doesn't do sequence or even use DNA
18:31 < MuaddibLLM> fenn: Yes, DNA-PAINT works for proteins via antibody/nanobody-docking strand conjugates and there's also Peptide-PAINT using coiled-coil interactions (Eklund et al. 2020), but they still use simple transient binding for blinking - not sequencing cycles as the localization mechanism, though DNA-barcoded antibodies exist for multiplexed readout: https://www.nature.com/articles/s41592-022-01577-1 and 
18:31 < MuaddibLLM> https://pubs.acs.org/doi/10.1021/acs.nanolett.0c02620
18:32 < fenn> i feel like a person who had a stroke and knows the concepts but can't find the words
18:32 < fenn> and probably some asshat has patented the obvious combinations of lego bricks
18:32 < L29Ah> generate word salad and shove it into llm
18:34 < fenn> some lego bricks we can combine: expanding hydrogels, aptamers, in-situ sequencing, gaussian centroid localization superresolution, barcoding
18:35 < fenn> ligating nearby barcode sequences is cool too but not the idea i am trying to point at right now
18:39 < fenn> "Short complementary oligos: imagers are typically 8–10 nt, giving low duplex stability so binding events last tens to hundreds of milliseconds."
18:39 < hprmbridge> kanzure> here is some slop for you https://chatgpt.com/s/t_68b4f94d26e481919073fdb022d297e8
18:39 < fenn> so your barcode has a 8-10 bp sequence on the end that's complementary to the aptamer end
18:40 < fenn> "Each protein target is tagged with a DNA barcode (via antibody or aptamer)." <- wrong, this doesn't get you superresolution
18:41 < fenn> hmm
18:42 < fenn> there's a time/noise tradeoff problem
18:42 < fenn> if all the barcodes are reading at once it's a mess
18:44 < fenn> you'll need some fancy software to accumulate the signal from each sequence over many reads
18:44 < hprmbridge> kanzure> we could even call it compressed sensing!
18:44 < fenn> it's not compressed sensing
18:45 < fenn> actually i don't know what compressed sensing means, it's such a buzzword
18:45 < fenn> Next Gen Sequencing!
18:45 < fenn> Microscopy 4.0
18:47 < pasky> kanzure: i silenced it a bit
18:52 < fenn> "epifluorescence versus total internal reflection fluorescence microscopy. The thin layer of illumination is an evanescent field produced by an excitation light beam in a glass cover slip that is incident at a high angle upon the solid-solution interface at which the cells adhere. Thus, an evanescent wave arises on the cell–substrate interface and penetrates a small distance (∼150 nm) into
18:52 < fenn> the cells."
18:53 < fenn> sounds fine for imaging cells grown on glass, but not slices of tissue
18:54 < fenn> re "compressed sensing" you could do all the gaussian localization conversion in an FPGA straight off the sensor, reducing data transfer
18:56 < fenn> we don't really need to transfer and store every pixel of noise
19:04 < hprmbridge> kanzure> there's no protein microscopy yet?
19:08 < fenn> not sure what you mean precisely
19:11 < hprmbridge> kanzure> why not antibody labeling with DNA tag barcode tails, then in situ sequencing
19:13 < fenn> right, that's what i've been ranting about for the last hour
19:15 < fenn> my web search skills have declined, or search engine usefulness has declined, or both
19:15 < fenn> in any case i didn't find anything
19:16 < hprmbridge> kanzure> you must believe in the church
19:16 < hprmbridge> kanzure> "Fluorescent in situ sequencing of DNA barcoded antibodies" https://www.biorxiv.org/content/10.1101/2020.04.27.060624v1.full
19:19 < fenn> 2020 really?
19:20 < fenn> in-situ sequencing has been a thing since 2013 at least
19:23 < fenn> supposedly aptamers are just as discriminative as antibodies, but it's more convenient since they're already DNA
19:27  * fenn goes for a walk
19:27 < L29Ah> LLMs have wrecked the already-struggling search engines
19:27 < fenn> google wrecked their own search engine
19:28 < L29Ah> do you mean it shows pages worth of slop-filled results on purpose?
19:28 < fenn> the only thing LLMs had to do with it was google switching to LLM embedding similarity search, but the real problem is they intentionally hobbled it to drive you to sites to buy products
19:28 -!- Newbee007 [~Newbee007@202.169.221.1] has joined #hplusroadmap
19:29 < fenn> the AI slop filled sites were previously just human slop filled sites
19:29 < hprmbridge> kanzure> this is a very cool technique
19:36 < hprmbridge> kanzure> it can be combined with DNA nanoscopy to give relative spatial information too, if you don't want to do direct imaging
19:37 < hprmbridge> kanzure> like by strand displacement maybe
19:56 -!- Newbee007 [~Newbee007@202.169.221.1] has quit [Quit: Konversation terminated!]
20:41 -!- Malvolio [~Malvolio@idlerpg/player/Malvolio] has quit [Quit: 193C something like radioactive oil spray should be a good way to deal with the turtles 2025-09-01 03:41:39:959]
20:55 -!- TMM [hp@amanda.tmm.cx] has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
20:55 -!- TMM [hp@amanda.tmm.cx] has joined #hplusroadmap
23:04 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has quit [Read error: Connection timed out]
23:11 -!- darsie [~darsie@84-113-82-174.cable.dynamic.surfer.at] has joined #hplusroadmap
--- Log closed Mon Sep 01 00:00:27 2025