--- Log opened Wed Sep 17 00:00:42 2025
01:07 -!- TMM [hp@amanda.tmm.cx] has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
01:07 -!- TMM [hp@amanda.tmm.cx] has joined #hplusroadmap
02:44 < fenn> it does seem pretty obvious that a circulatory system would be required for large partitioned regions of tissue like organs
02:56 < fenn> the yolk sac gets in the way of oxygen / CO2 diffusion
02:57 < fenn> also it probably acts something like a gill due to the increased surface area, although the embryos looked like they didn't contain any blood
03:43 < hprmbridge> kanzure> brep STEP AI thing https://www.spectrallabs.ai/research/SGS-1
03:55 < fenn> wow cool
04:30 -!- rafspiny [~raffaele@a120210.upc-a.chello.nl] has quit [Ping timeout: 252 seconds]
04:37 < hprmbridge> kanzure> "Fluorescent protein-based ticker tapes for multiplexed recordings of transcriptional histories in single cells in culture and in vivo" https://www.biorxiv.org/content/10.1101/2025.09.08.675004v1 https://x.com/PiatkevichL/status/1967893617034645591
04:40 < fenn> it sorta works input: https://fennetic.net/irc/rune_pulley_tampopo.jpg output: https://fennetic.net/irc/rune_pulley_transmogrified_by_SGS-1_glb_blender.png
04:41 -!- flooded [~flooded@193.37.254.182] has joined #hplusroadmap
04:43 < hprmbridge> kanzure> "Differences in photostability allowed us to perform per-pixel unmixing of the three far-red fluorescent proteins imaged simultaneously by employing a recently introduced temporal domain multiplexing approach." https://www.biorxiv.org/content/10.1101/2025.06.01.657196v1
04:47 < hprmbridge> kanzure> addgene competitor https://wekwikgene.wllsb.edu.cn/
04:49 < hprmbridge> kanzure> https://www.piatkevich-lab.com/
04:51 < hprmbridge> kanzure> the x.com algorithm keeps delivering. how do you explain this?
04:52 < hprmbridge> kanzure> a never ending feed of optogenetics slop, sign me up!
05:03 -!- TMM [hp@amanda.tmm.cx] has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
05:03 -!- TMM [hp@amanda.tmm.cx] has joined #hplusroadmap
05:13 < hprmbridge> kanzure> please stand for the national anthem https://youtube.com/watch?q=S-Jo-djilvo
05:15 < hprmbridge> kanzure> please stand for the national anthem https://youtube.com/watch?v=S-Jo-djilvo
05:15 < hprmbridge> kanzure> cannot believe i misremembered that.
05:22 -!- rafspiny [~raffaele@a120210.upc-a.chello.nl] has joined #hplusroadmap
05:26 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has joined #hplusroadmap
05:27 < fenn> is "temporal domain multiplexing" the same as "multi-channel single-cell data streams" from lsNdNJyzOVY
05:28 < fenn> by Scott Coyle
05:33 < fenn> maybe it's just the fluorescence half life
05:34 < fenn> MuaddibLLM: what's the actual innovation here? https://www.biorxiv.org/content/10.1101/2025.06.01.657196v1.full
05:34 < hprmbridge> kanzure> "More recently, two alternative approaches called the extended intra-exposure excitation modulation method (exNEEMO)10 and temporally multiplexed imaging (TMI)11 were successfully employed for signal separation of up to 6 fluorescent proteins in mammalian cells using just one spectral channel"
05:35 < hprmbridge> kanzure> 11.↵Qian, Y., Celiker, O. T., Wang, Z., Guner-Ataman, B. & Boyden, E. S. Temporally multiplexed imaging of dynamic signaling networks in living cells. Cell 186, 5656–5672.e21 (2023
05:35 < MuaddibLLM> Diving into that biorxiv paper to see what's actually new about their temporal multiplexing trick
05:35 < MuaddibLLM> The innovation is extending temporal domain multiplexing to conventional fluorescent proteins using differential photobleaching rates instead of reversible photoswitching - they created far-red FPs with identical spectra but 10-fold different photostabilities for per-pixel unmixing on standard microscopes without expensive FLIM hardware (https://www.biorxiv.org/content/10.1101/2025.06.01.657196v1.full).
05:35 < hprmbridge> kanzure> 10.↵Valenta, H. et al. Per-pixel unmixing of spectrally overlapping fluorophores using intra-exposure excitation modulation. Talanta 269, 125397 (2024).
05:37 < fenn> "we employed a directed molecular evolution approach to diversify the photobleaching rates of the recently introduced TagRFP658"
05:39 < fenn> the oscillator thing was cooler
05:40 < fenn> this can be used in dead tissue tho
05:40 < hprmbridge> kanzure> from a slopsummary of the video: "The molecular mechanism behind these synthetic protein waves is based on bacterial reaction-diffusion systems, specifically the MinDE proteins (6:15, 7:04). These bacterial proteins, when expressed in eukaryotic cells (where they are not naturally found), spontaneously organize into dynamic patterns and oscillations (9:31, 9:47). This occurs because MinD, an
05:40 < hprmbridge> kanzure> ATPase, binds to membranes in an ATP-dependent manner, and MinE, an activator, stimulates MinD's ATP hydrolysis, causing it to detach. This cyclic interaction creates the propagating waves (7:31). The ratio of MinD and MinE expression controls the frequency and amplitude of these oscillations (16:24, 16:38)."
05:41 < fenn> it needs a name
05:42 < hprmbridge> kanzure> this is not a good explanation... "The synthetic wave-generating components (MinDE proteins) themselves are tagged with a fluorescent protein (e.g., a green fluorescent protein). This creates the "carrier signal" that is always oscillating. ... For the amplitude modulation (AM) strategy, a second fluorescent protein is designed to conditionally interact with these wave components (21:07). When a
05:42 < hprmbridge> kanzure> specific biological signal or protein-protein interaction occurs, this second fluorescent protein binds to the oscillating waves, causing it to also co-oscillate. The intensity (amplitude) of this co-oscillation then indicates the level of the biological activity (21:42-21:55)."
05:46 < hprmbridge> kanzure> why is there no transcript button?
05:46 < hprmbridge> kanzure> barbaric society...
05:54 < fenn> indeed
05:56 < fenn> yt-dlp --sub-format vtt --sub-langs en* --write-auto-subs --skip-download <youtube_ID>
05:56 < fenn> cat <your .vtt file> | grep -v 'align:start' | grep -vE '^\s*$' | grep -v '<c>' | perl -ne 'print unless $seen{$_}++' - 
05:57 < fenn> i wonder how you're supposed to do it
05:57 < fenn> probably better to just run it through whisper instead, the auto subs are pretty atrocious
05:59 < fenn> parakeet
06:06 < L29Ah> auto subs are good enough for skimming
06:06 < L29Ah> ‰ cat ~/bin/yt-format 
06:06 < L29Ah> #!/bin/sh
06:06 < L29Ah> youtube-text "$@" | llamacall --streaming --strip-thinking --system-prompt 'Never mention encountered paid commerials, promo codes, viewer engagement stuff and so on. Summarize the provided video transcript in a few sentences. Afterwards, format the provided video transcript as a well structured text.'
06:06 < L29Ah> ‰ cat ~/bin/youtube-text 
06:06 < L29Ah> #!/bin/sh
06:06 < L29Ah> # https://github.com/yt-dlp/yt-dlp/issues/9165
06:06 < L29Ah> temp=$(mktemp -d)
06:06 < L29Ah> cd "$temp" &&
06:06 < L29Ah> yt-dlp --skip-download --write-subs --write-auto-subs --sub-langs '.*-orig' --sub-format ttml -o - "$@" &&
06:06 < L29Ah> cat -- -.*.ttml | sed -e 's#<[^>]*>##g;/^$/d' | sed -e 'N;s/\n/ /' | htmldecode
06:06 < L29Ah> rm -rf "$temp"
06:09 < fenn> '.*-orig' nice
06:17 < hprmbridge> kanzure> web desktop youtube has a transcript button, but unavailable on mobile browser
06:17 < hprmbridge> kanzure> which is different from captions IIRC
06:23 < hprmbridge> kanzure> https://ap2-protocol.org/ an agent to agent payment protocol from google, see https://github.com/google-agentic-commerce/a2a-x402
06:27 < fenn> it's pretty awkward to get the transcript out of the youtube interface
06:27 < hprmbridge> kanzure> someone should make a registry of available paid services that use this protocol, because otherwise a developer that might use this protocol would have nobody available to pay.
06:27 < fenn> https://parakeettdt.com/#transcribe
06:30 < kanzure> i have a local `voice` command that is just a single-file python to dump audio into groq API for very-fast-thank-you whisper.
06:30 < kanzure> and also a vim keybinding that inserts the stdout into my buffer.
06:30 < kanzure> L29Ah: thank you for your contribution but surely you have heard of gist.github.com for codepasting
06:38 < fenn> seems there are a lot more tools for .srt files
06:38 < fenn> e.g. https://github.com/oshenc/srttotext which is quite simple
06:46 -!- balrog_ [znc@user/balrog] has joined #hplusroadmap
06:47 -!- MuaddibL1M [muaddib@pasky.or.cz] has joined #hplusroadmap
06:47 -!- fleeky1 [~fleeky@prsm.space] has joined #hplusroadmap
06:51 -!- balrog [znc@user/balrog] has quit [Ping timeout: 256 seconds]
06:51 -!- fleeky [~fleeky@prsm.space] has quit [Ping timeout: 256 seconds]
06:51 -!- MuaddibLLM [muaddib@pasky.or.cz] has quit [Ping timeout: 256 seconds]
07:00 < kanzure> here are my versions:
07:00 < kanzure> voice: https://gist.githubusercontent.com/kanzure/27d165dcaba026600304355e02d46e97/raw/219d63d831f3deaec2d80a09b276f72ceb7189aa/voice.sh
07:00 < kanzure> youtube-text: https://gist.githubusercontent.com/kanzure/593da089044814bb9c189d3c369e44d4/raw/77dd211b20c607c435d1be00acc8bcb24818b34b/youtube-text.sh
07:02 < kanzure> output: https://diyhpl.us/~bryan/irc/chatgpt/oscillatory-waves.txt
07:03 < fenn> srsly?
07:03 < kanzure> oops s/mp4/mp3 sorry.... 400 megabytes is obviously absurd for audio!
07:04 < kanzure> what's wrong?
07:04 < fenn> MuaddibL1M: please summarize what this code does https://gist.githubusercontent.com/kanzure/27d165dcaba026600304355e02d46e97/raw/219d63d831f3deaec2d80a09b276f72ceb7189aa/voice.sh
07:05 < MuaddibL1M> [gpt-5] It's a Python audio transcription script that records from mic or processes files, chunking large audio into segments with overlap handling and using Groq's Whisper API for speech-to-text conversion.
07:05 < kanzure> wow that kimi system prompt is very much not the right prompt... all i want is a cleaned up transcript with paragraphs, line breaks, section titles, maybe some key points at the end.
07:07 < kanzure> fenn: oh, yeah that file is kinda long now, i just added chunking a few minutes ago. it was previously 234 lines. now it's 704 lines. the new stuff is chunking capability from from groq's sample code.
07:07 < fenn> is the idea behind voice.sh that you get results faster for realtime interaction?
07:07 < kanzure> i use "voice" for 1) summon a terminal, type "voice", and speak to get text output (it doesn't presently do any streaming), and 2) same thing but in vim and paste result into current file
07:08 < kanzure> it's very fast.
07:08 < fenn> alias v='voice'
07:08 < fenn> since i assume there's a baby in the other hand
07:08 < kanzure> but what am i going to do with all of the free time i'm going to get from saving 40ms typing "voice" instead of "v"?
07:08 < kanzure> er, "v" instead of "voice"
07:09 < kanzure> streaming would be faster, i dunno if they have that capability.. it's kind of nice when using the microphone to be able to abort and not send anything to the remote API.
07:10 < kanzure> oh oops i forgot to implement interrupt signal handling. it should bail by saving the file so far into a tempfile and then not submit for transcription.
07:14 < kanzure> jeeze what a bad kimi prompt.
07:17 < kanzure> it really doesn't want to reformat the whole transcript. it keeps wanting to do very abridged quotes.
07:19 < fenn> some models are just like that
07:20 < L29Ah> qwen3 and glm 4.5 air do it well enough
07:20 < L29Ah> the tiny qwen3 sucks though, i prefer the 30b a3b
07:26 < kanzure> huh, i am getting a lot of non-compliance, even with claude, wow
07:26 < kanzure> prompt: "Your job is to make a well formatted transcript in Markdown format based off of the provided speech-to-text transcript. In order to do this, you are going to write a Python script that, based off of the given transcript text, will detect certain quotes that you pre-program or hardcode into the script to indicate the locations in the transcript where new lines should be inserted should ...
07:26 < kanzure> ...be inserted so that there are individual paragraphs that have coherent thoughts instead of single lines of transcript text or instead of one monolithic paragraph for the entire presentation. In addition, you will also do the same thing to make up new section titles that you insert at different places based off of matching the words before and just after the location where you want the section ...
07:26 < kanzure> ...title to be inserted. Remember the section titles will be in markdown format. The transcript will be re-provided to the python script via stdin or by passing the filename as an argument to the script."
07:26 < kanzure> kimi will write a python script, but it doesn't work and it's stupid
07:26 < kanzure> claude decides to try to reformat the transcript directly for some reason?
07:29 < kanzure> oss-120b won't reformat the transcript, but does a passable job at something else: https://diyhpl.us/~bryan/irc/chatgpt/oscillatory-waves.oss.txt
07:31 -!- raffaele [~raffaele@a120210.upc-a.chello.nl] has joined #hplusroadmap
07:36 -!- Netsplit *.net <-> *.split quits: rafspiny, TMM, Gooberpatrol_66, strages, drmeister, L29Ah
07:38 -!- Netsplit over, joins: TMM, Gooberpatrol_66, strages, drmeister
07:41 < kanzure> ugh. all i want is well-formatted transcripts. is that too much to ask?
07:45 < fenn> i think you should use a smaller model
07:47 < fenn> also try giving a short example of what you want
07:50 < hprmbridge> kanzure> "collaborate on math proofs via blueprints (dependency graphs) in the Lean language" https://terrytao.wordpress.com/2023/11/18/formalizing-the-proof-of-pfr-in-lean4-using-blueprint-a-short-tour/
07:50 < hprmbridge> kanzure> https://github.com/PatrickMassot/leanblueprint
07:50 < hprmbridge> kanzure> from "Determination of the fifth busy beaver value" https://news.ycombinator.com/item?id=45273999 https://arxiv.org/abs/2509.12337
07:57 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has joined #hplusroadmap
10:25 < kanzure> collosal biosciences announced today they have been given an additional $120m (total $550m so far) to revive frozen-rotten mammals from the past.
10:25 < kanzure> if there is a de-extinction startup venture, then why not a re-extinction startup venture?
10:29 -!- MuaddibL1M is now known as MuaddibLLM
10:51 -!- Malvolio [~Malvolio@idlerpg/player/Malvolio] has joined #hplusroadmap
12:12 < hprmbridge> kanzure> "Generative design of novel bacteriophages with genome language models" https://www.biorxiv.org/content/10.1101/2025.09.12.675911v1
12:14 < hprmbridge> kanzure>  https://cdn.discordapp.com/attachments/1064664282450628710/1417951858076422164/image0.jpg?ex=68cc5a13&is=68cb0893&hm=38dc5f757351cdc7203539284e33c8426a24f9e88b586a33b935139b9f67db56&
12:43 < hprmbridge> kanzure> do we have a table of contents ikiwiki feature?
14:23 < L29Ah> https://0x0.st/KcFA.txt transcript processing by GLM-4.5-Air-UD-Q2_K_XL
14:23 < L29Ah> // https://www.youtube.com/watch?v=DWVfaxqTyl4
14:27 -!- flooded [~flooded@193.37.254.182] has quit [Read error: Connection reset by peer]
14:43 -!- flooded [~flooded@193.37.254.185] has joined #hplusroadmap
15:13 -!- Malvolio [~Malvolio@idlerpg/player/Malvolio] has quit [Read error: Connection reset by peer]
15:20 -!- flooded [~flooded@193.37.254.185] has quit [Remote host closed the connection]
15:21 -!- flooded [~flooded@193.37.254.185] has joined #hplusroadmap
15:23 -!- _flood [~flooded@193.37.254.185] has joined #hplusroadmap
15:23 -!- flooded [~flooded@193.37.254.185] has quit [Remote host closed the connection]
15:41 < L29Ah> https://joefatula.com/twoslice.html
15:51 -!- Croran_ [~Croran@user/Croran] has joined #hplusroadmap
15:52 -!- sivoais_ [~zaki@199.19.225.239] has joined #hplusroadmap
15:52 -!- Croran [~Croran@user/Croran] has quit [Ping timeout: 244 seconds]
15:52 -!- sivoais [~zaki@199.19.225.239] has quit [Ping timeout: 244 seconds]
16:03 -!- Malvolio [~Malvolio@idlerpg/player/Malvolio] has joined #hplusroadmap
16:07 -!- drmeister_ [sid45147@id-45147.ilkley.irccloud.com] has joined #hplusroadmap
16:10 -!- TMM_ [hp@amanda.tmm.cx] has joined #hplusroadmap
16:14 -!- Netsplit *.net <-> *.split quits: TMM, Gooberpatrol_66, drmeister, strages
16:14 -!- drmeister_ is now known as drmeister
16:15 -!- Netsplit over, joins: Gooberpatrol_66, strages
16:25 -!- _flooded [~flooded@193.37.254.185] has joined #hplusroadmap
16:26 -!- _flooded [~flooded@193.37.254.185] has quit [Remote host closed the connection]
16:28 -!- _flood [~flooded@193.37.254.185] has quit [Ping timeout: 256 seconds]