--- Log opened Wed Jul 02 00:00:29 2025
00:47 < hprmbridge> kanzure> https://www.zach.be/p/why-is-it-so-hard-for-startups-to
00:47 < hprmbridge> kanzure> er, that's an article about Cadence
00:55 < hprmbridge> Katylase> really?
00:56 < hprmbridge> Katylase> that's so cool!
04:30 -!- justanotheruser [~justanoth@gateway/tor-sasl/justanotheruser] has quit [Remote host closed the connection]
04:31 -!- justanotheruser [~justanoth@gateway/tor-sasl/justanotheruser] has joined #hplusroadmap
05:53 -!- 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.]
05:53 -!- darsie [~darsie@84-113-82-174.cable.dynamic.surfer.at] has joined #hplusroadmap
07:03 < kanzure> yes on mathchan there were math problem captchas before posting.. or something.
07:10 -!- TMM [hp@amanda.tmm.cx] has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
07:10 -!- TMM [hp@amanda.tmm.cx] has joined #hplusroadmap
07:51 -!- gl00ten [~gl00ten@2001:8a0:7ee5:7800:46d9:f5c:17a2:432] has joined #hplusroadmap
07:52 < RangerMauve> @Katylase what is your game?
08:02 < kanzure> https://dood.al/pinktrombone/
08:03 < RangerMauve> 🤪
08:05 < kanzure> welp that's definitely weird
08:05 < kanzure> who would make this sort of thing and not also hook up a phonetic alphabet?
08:06 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has left #hplusroadmap [Disconnected: closed]
08:08 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has joined #hplusroadmap
08:11 < kanzure> https://vadosware.io/post/why-dont-more-people-use-throat-mics/
08:14 < kanzure> https://news.ycombinator.com/item?id=31931887
08:14 < kanzure> according to military microphone vendors you should be able to use throat microphones to pick up nearly inaudible whispered human speech, but i see no evidence of this on youtube in any of the reviews (nor anyone trying one way or the other)
08:15 < kanzure> subvocalization tech is cool but if nearly-inaudible-vocalization microphones exist and work and can convey ~inaudible human speech then that is very useful
08:16 < kanzure> use case: smartphone text entry is horrible, everyone should be voice-only input into phones now with modern speech recognition technologies, but often you can't freely vocalize outloud when you are on the go or out in public (for those of us who venture in public!)
08:17 < kanzure> some of the stormtrooper cosplay people use throat microphones for voice amplification from their costumes but their youtube reviews don't cover whispering because they are explicitly trying to be heard through their costume so they don't think about testing whispering
08:19 < kanzure> nobody calls it a laryngophone?
08:23 < L29Ah> you can pick up throat sounds but it is lips that make it speech
08:24 < kanzure> "throat sounds aren't speech! NOOOO!" okay chomsky
08:25 < RangerMauve> Yeah I dove into that rabbit hole a while ago and it all came up lacking. Even the fancier setups still had shit audio quality
08:26 < RangerMauve> THe subvocalizing stuff wasn't great either because it needed to be trained on wordlists IIRC
08:26 < kanzure> i'm willing to forego subvocalization, but also if i was interested in subvocalization i think i would have to discard a lot of prior subvocal speech recognition research given our new glorious era of speech recognition
09:03 -!- gl00ten [~gl00ten@2001:8a0:7ee5:7800:46d9:f5c:17a2:432] has quit [Ping timeout: 272 seconds]
09:07 < hprmbridge> kanzure> "wishlist of plausible DNA enzymes" https://x.com/SynBio1/status/1939668234216587449
09:18 < L29Ah> nature gave you the perfect output device: your fingers
09:18 < L29Ah> no need to abuse windpipes and feeding orifices
09:18 < hprmbridge> kanzure> yes but phone haptic interfaces are slooooow
09:20 < L29Ah> use a chorded keyboard
09:20 < hprmbridge> kanzure> also my hands are not free because I'm doing things..if I was free to type I would just use an actual computer and keyboard.
09:51 < kanzure> it might be possible to "estimate" the "size of the global human population" as understood by LLMs by trying to extract how many people (names of people) it knows or has in its weights. One way I was thinking of doing this is doing city-by-city querying ("Name 100 people that live in London") and then randomly sampling the provided list, doing a web search to validate that those people live in ...
09:51 < kanzure> ...the city. And then coming up with an estimate of how many people the model "knows".
09:54 < kanzure> of course, the results will change with prompting and temperature, so i'm not sure the results would be interesting? plus the outcome ("a very low fraction of the human population") is basically guaranteed; not sure there is any value to knowing an actual number here?
09:56 < kanzure> generative mentionability is going to be a finite and scarce resource and estimating how scarce the resource is would be somewhat financially valuable
10:00 < hprmbridge> Katylase> my game? the one im making? its somewhat like a combination of animal crossing with foldit😅 a dialogue based one with protein npcs...
10:06 < kanzure> uhhh did anything ever come of the neuralink olfactory bulb implant demo? pretty sure that is the highest bandwidth, highest molecular-target-discriminator molecular sensor device we have ever created. dogs are supposed to discriminate between millions of odors or something, pretty sure pigs are similar.
10:09 < kanzure> "1/8 of a dog’s brain is devoted to analyzing smells" "canine olfactory bulb is 40x larger than in humans (relative to total brain size)" 
10:09 < kanzure> "A bloodhound can detect 1 part per 10 trillion, roughly equivalent to detecting a single drop of blood diluted in 20 Olympic-sized swimming pools."
10:09 < kanzure> "German Shepherds have been trained to detect explosives at 1–2 parts per billion"
10:12 < kanzure> uhh yeah i must be missing something. if you could get reliable signal out of the olfactory cortex then you can decouple pig odor detection capabilities from pig behavior such as tracking or signaling (like oinking loudly to your handler)
10:13 < kanzure> this is a generalized molecular sensor. everyone should have one of these in their homes, labs, at work, etc.
10:21 < RangerMauve> https://www.cell.com/cell/abstract/S0092-8674(25)00454-4 gotta get me some of these so I can get better night vision >:)
10:25 < kanzure> if the canine olfactory limit is one part per trillion for certain molecular targets, then the operational scent threshold limit might be closer to one part per billion or worse depending on dog training, dog/handler teamwork efficiency, distractions, etc.
10:27 < kanzure> this seems like an easy one or two orders of magnitude performance improvement on scent tracking tasks if by using neuralink you can decouple signal detection from animal behavior (such as reporting, trail following, etc)
10:39 < kanzure> also: it should be possible to do environmental sampling (either atmospheric sampling or ground/dirt sampling) to store and preserve odors in a more neutral environment for later scent detection by animal. you can collect a lot of samples very quickly, much faster than you can have animals cover a whole search surface area. you can grid-approximate the samples and map back detection strengths ...
10:39 < kanzure> ...or results.
11:16 < kanzure> ok the requirement to imprint, and the likely unique coding per odorant in the olfactory pathways, minimizes the utility a little bit. if there was a universal olfactory signal encoding scheme between each dog or pig then that would be great but seems very unlikely.
11:26 < kanzure> "The LLM can develop much faster than I can build a mental model. It's very easy to get to a point where you don't know what's going on, a bunch of bugs have been introduced and you can't easily fix them or refactor because you're essentially the new guy on your own project."
11:53 < kanzure> odd take, not necessarily wrong: "I think (futurology / science fiction) that they will make some kind of brain link, but there won't be any translations happening in between, just raw brain signals from one to the other, like an extra sensory input; there won't be any encoding or data that can be translated to speech or images, but the connected brains will be able to learn to comprehend and ...
11:53 < kanzure> ...send the signals to / from each other and learn to communicate that way." 
11:56 < kanzure> "An instantaneous voice-synthesis neuroprosthesis" https://www.nature.com/articles/s41586-025-09127-3 ... "Here we demonstrate a brain-to-voice neuroprosthesis that instantaneously synthesizes voice with closed-loop audio feedback by decoding neural activity from 256 microelectrodes implanted into the ventral precentral gyrus of a man with amyotrophic lateral sclerosis and severe dysarthria."
11:57 < kanzure> "Paradromics, a BCI-focused startup based in Austin, Texas, wants to go ahead with clinical trials of a speech neural prosthesis and is already seeking FDA approval. “They have a 1,600 electrode system, and they publicly stated they are going to do speech,” Stavisky says. “David Brandman, our co-author, is going to be the lead principal investigator for these trials, and we’re going to ...
11:57 < kanzure> ...do it here at UC Davis.”
11:59 < kanzure> apparently ventral precentral gyrus controls vocal tract muscles.
12:00 < kanzure> that particular version had a very high word error rate on intelligibility (>10%, apparently >40% WER) but i would guess number of microelectrodes should be increased, and more data science magic applied to signal processing, possibly also next token prediction
13:14 < hprmbridge> norkatron> https://huggingface.co/tomg-group-umd/huginn-0125
13:14 < hprmbridge> norkatron> https://www.arxiv.org/abs/2502.05171
13:14 < hprmbridge> norkatron>
13:14 < hprmbridge> norkatron> reasoning in latent space , no conversion , kinda what you were talking about with brain to brain links ?
13:16 < hprmbridge> norkatron> i suppose the real question is would we consider the electrical activity in the brain a type of latent space or is that a metaphor failure ..
13:26 < kanzure> thalamocortical bridges between conjoined twins have been known to relay perceptual data such as vision
13:58 -!- superkuh [~superkuh@user/superkuh] has quit [Ping timeout: 260 seconds]
14:22 < L29Ah> https://arxiv.org/html/2506.22405v1
14:51 < hprmbridge> kanzure> what professions are providing that 20% rate, is that general physician population? or academic med school professors?
14:51 < hprmbridge> kanzure> 'generalist physicians'
15:23 -!- darsie [~darsie@84-113-82-174.cable.dynamic.surfer.at] has quit [Ping timeout: 260 seconds]
15:48 -!- stipa_ [~stipa@user/stipa] has joined #hplusroadmap
15:52 -!- stipa [~stipa@user/stipa] has quit [Ping timeout: 260 seconds]
15:52 -!- stipa_ is now known as stipa
16:05 -!- gl00ten [~gl00ten@2001:8a0:7ee5:7800:46d9:f5c:17a2:432] has joined #hplusroadmap
16:10 -!- gl00ten [~gl00ten@2001:8a0:7ee5:7800:46d9:f5c:17a2:432] has quit [Quit: Leaving]
17:27 -!- superkuh [~superkuh@user/superkuh] has joined #hplusroadmap
17:32 < hprmbridge> Eli> For the Microsoft ai paper? “we also evaluated 21 practicing physicians from the US and UK, each with 5-20 years of clinical experience. On the same tasks, these experts achieved a mean accuracy of 20% across completed cases. “
22:09 -!- darsie [~darsie@84-113-82-174.cable.dynamic.surfer.at] has joined #hplusroadmap
22:39 < fenn> this is not really what i meant by a blender CAD program but i guess it's better than nothing: https://www.youtube.com/watch?v=Ooh05WF__80
22:39 < fenn> https://bonsaibim.org
22:55 < fenn> 3d cad standard parts library https://boltsparts.github.io/
22:59 < hprmbridge> lordkek__> TRT (Temporal Resonance Technology)
23:00 < fenn> go away
23:00 < fenn> this is a no-schizopoasting zone
23:27 < fenn> what are they going to call the new process nodes after "1nm"? since they aren't really feature sizes... "-1nm"?
23:37 -!- TMM [hp@amanda.tmm.cx] has quit [Quit: https://quassel-irc.org - Chat comfortably. Anywhere.]
23:37 -!- TMM [hp@amanda.tmm.cx] has joined #hplusroadmap
23:39 < fenn> i don't remember a neuralink olfactory implant but koniku is making an olfactory neuron tissue culture on chip device for detecting chemicals in the environment, starting with explosives nerve gas etc
23:43 < fenn> there could be a universal olfactory signal encoding scheme, it's been shown to exist with other types of data https://arxiv.org/abs/2505.12540
23:44 < fenn> i expect you to have to do some map-mapping to fit the representations to one another, but then it should line up
23:44 < fenn> (complete speculation)
23:48 < hprmbridge> nmz787> https://youtu.be/Cw7VC6YjF6I
23:48 < hprmbridge> nmz787> currently $24/year with some july coupon on their site
23:48 < hprmbridge> nmz787> it looks nice
23:49 < hprmbridge> nmz787> it's cheap in terms of software that could be short lived... and even cheaper if it made experimenting easier... but all the parametrics seem locked in their fileformat
23:50 < hprmbridge> nmz787> I think I will try to see if my kid can tolerate it long enough to learn to actually use it in a creative way
23:50 < hprmbridge> nmz787> I have been thinking a lot about forestry and ag mini robots
23:51 < fenn> .t
23:51 < saxo> First Look - 3D Creator
23:51 < fenn> solidworks for hobbyists?
23:54 < fenn> solidworks in the cloud
23:55 < fenn> if you can export it to solidworks desktop then you should be able to use cracked software in perpetuity if needed
23:56 < fenn> you have to wonder what leads to statements like "contact your local value added reseller"
23:56 < hprmbridge> nmz787> like I want a robot that will carry a gallon of water or something, and water all the seedlings if I am feeling lazy to get water hoses setup. I want a mini tree-spade for weeding on a robot... something like https://youtu.be/F8AkuD6CCFY?t=45   or https://youtu.be/EXxFRQ3URZg?t=115
23:56 < hprmbridge> nmz787> and also a robot with a 4" chainsaw on it (basically a battery drill worth of motor and gears)
23:57 < hprmbridge> nmz787> yeah reseller, meanwhile I got the vid from their product "buy it now" page
23:57 < hprmbridge> nmz787> (and by mini tree spade, I was thinking like an inch at max diameter, or less)
23:58 < hprmbridge> nmz787> https://www.solidworks.com/solution/solidworks-makers was where I found the $24/year thing
23:59 < hprmbridge> nmz787> they also have a windows option if you don't like cloud
--- Log closed Thu Jul 03 00:00:30 2025