--- Log opened Fri Sep 22 00:00:29 2023
01:23 -!- Croran [~Croran@user/Croran] has quit [Ping timeout: 264 seconds]
01:43 -!- darsie [~darsie@84-113-55-200.cable.dynamic.surfer.at] has joined #hplusroadmap
02:20 -!- Croran [~Croran@user/Croran] has joined #hplusroadmap
02:53 -!- sphertext_ [~sphertext@user/sphertext] has joined #hplusroadmap
02:56 -!- sphertext [~sphertext@user/sphertext] has quit [Ping timeout: 264 seconds]
03:14 -!- Croran [~Croran@user/Croran] has quit [Ping timeout: 244 seconds]
03:16 -!- Croran [~Croran@user/Croran] has joined #hplusroadmap
04:04 < hprmbridge> Eli> Ok, I’m probably not as knowledgeable as I need you to be here.
04:10 -!- ike8 [12fdf2ee08@irc.cheogram.com] has quit [Ping timeout: 244 seconds]
04:11 < hprmbridge> Eli> So, what is the proper dose response? What are the adverse effects? There appears to be no large RCTs on this. There are a bajillion compounds that never made it past phase 3 trials but looked great previously. Some compounds even make it past phase 3 and then need a recall. I don’t like to simp for the fda, but we’re really glossing over things if we give stuff to seemingly healthy people to help
04:11 < hprmbridge> Eli> them live longer without knowing dose response or adverse effects.
04:18 < hprmbridge> Eli> If you’re referring to the beta carotene studies in the 90s, the issue was that smokers and people with asbestos exposure were dying prematurely. This is probably not because the beta carotene was causing cancer so much as it is likely that people with tumors already existing had much more rapid growth of tumors after taking beta carotene. If excess exogenous antioxidant were pro cancer we would
04:18 < hprmbridge> Eli> probably see warnings about fruit and vegetables ingestion. Instead, we see the opposite.
04:21 < hprmbridge> Eli> The main concern is taking antioxidants when you currently have cancer or have had cancer and lingering cancer cells may be in your body.
05:01 -!- test_ [flooded@gateway/vpn/protonvpn/flood/x-43489060] has joined #hplusroadmap
05:05 -!- test__ [flooded@gateway/vpn/protonvpn/flood/x-43489060] has quit [Ping timeout: 240 seconds]
05:05 -!- test__ [flooded@gateway/vpn/protonvpn/flood/x-43489060] has joined #hplusroadmap
05:05 -!- flooded [flooded@gateway/vpn/protonvpn/flood/x-43489060] has quit [Ping timeout: 244 seconds]
05:06 -!- flooded [flooded@gateway/vpn/protonvpn/flood/x-43489060] has joined #hplusroadmap
05:07 -!- test_ [flooded@gateway/vpn/protonvpn/flood/x-43489060] has quit [Ping timeout: 240 seconds]
05:44 -!- yashgaroth [~ffffffff@2601:5c4:c780:6aa0:589c:394f:ddcd:f875] has joined #hplusroadmap
06:40 -!- test__ is now known as _flood
07:13 -!- ike8 [12fdf2ee08@irc.cheogram.com] has joined #hplusroadmap
07:43 < ike8> The Sheekey Science Show on YouTube has a few excellent videos on Cellular Senescence.
08:10 -!- test_ [flooded@gateway/vpn/protonvpn/flood/x-43489060] has joined #hplusroadmap
08:10 -!- test__ [flooded@gateway/vpn/protonvpn/flood/x-43489060] has joined #hplusroadmap
08:14 -!- flooded [flooded@gateway/vpn/protonvpn/flood/x-43489060] has quit [Ping timeout: 264 seconds]
08:14 -!- _flood [flooded@gateway/vpn/protonvpn/flood/x-43489060] has quit [Ping timeout: 264 seconds]
09:29 < hprmbridge> soul_syrup> What would be your opinions(not ethical opinions) on using an engineered "animal" for synthetic cerebral engineering? since a lot of the cerebral organoid research is conceptually in the direction of synthetic biology, why dont we just route the senses of an animal into a nice vr env,  and use this animal as storage for the synthetic brain and for nutrient/waste care? I know it sounds far out, but
09:29 < hprmbridge> soul_syrup> it does follow the line of thought that we already have biological systems that perform what we are trying to do artificially
09:30 < hprmbridge> soul_syrup> this would provide a much better env, vascular structure etc for the synth brain
10:13 < fenn> we have talked about breeding frogs for this purpose, since they aren't regulated to oblivion like mammals
10:14 < fenn> afaik there isn't anything special about the human chassis, but the glial cells are high performance, and probably the neurons too
10:15 < fenn> so if you really want to get good performance from you froggies, they have to have human brains, and then the ethics people come piling in
10:18 < fenn> i don't think a VR headset will be adequate bandwidth or specific enough for what i'd want, which is effectively expanding my own brain by interpolation
10:19 < fenn> you could imagine continuing the cerebral growth program virtually by linking cells that are close together in the human brain with cells that are close together in the froggie brain
10:19 < fenn> there's also the problem of what happens when the frog dies
10:20 < fenn> for doing OCR or fraud detection or bombing brown people in deserts or whatever, sure. it would be more energy efficient and cheaper
10:25 < fenn> i was peripherally involved in some zebrafish brain light sheet microscopy to sense neuronal activations across the entire brain with high spatial and temporal resolution, and it was not cheap
10:25 < fenn> we called it "the fish matrix"
10:26 < fenn> there were LCD screens covering the fish's field of view, and the fish itself was suspended in a narrow channel of water between two precisely gelatin blocks that fit around its body
10:30 < fenn> since the gelatin, fish, and water are all optically clear, we can excite the brain from the side with blue light, which stimulates fluorescent calcium sensing receptors or voltage sensitive dyes
10:30 < fenn> unfortunately blue light is cytotoxic
10:31 < fenn> perhaps a more highly genetically engineered system could emit light itself, but then you would have to disambiguate the Z-axis with plenoptic light field microscopy
10:32 < fenn> so if you don't want to turn a computing problem into an optics problem, you need to have good long term brain tissue electrodes
10:39 < fenn> https://static-content.springer.com/esm/art%3A10.1038%2Fnmeth.2434/MediaObjects/41592_2013_BFnmeth2434_MOESM207_ESM.avi https://static-content.springer.com/esm/art%3A10.1038%2Fnmeth.2434/MediaObjects/41592_2013_BFnmeth2434_MOESM204_ESM.avi
10:41 < fenn> even with fancy hamamatsu scientific cameras and high numerical aperture, you still don't get enough signal to noise at high enough scan rates to resolve individual neuron firings across the entire brain. a couple orders of magnitude more and it would be a solved problem
10:43 < fenn> some of this could be dealt with by growing a flat pancake brain, to simplify the optics problem
10:44 < kanzure> was the VR headset comment part of this conversation or a different one, and could you explain the leap to why VR headset is relevant here
10:44 < fenn> we don't need to image *every* neuron of course, but where do you find the output?
10:44 < fenn> a fish in clear gelatin looking at images on LCD screens is equivalent to a VR headset
10:45 < fenn> VR was mentioned in soul_syrup's original question
10:46 < kanzure> oh weird
10:47 < kanzure> just put them in front of a normal computer screen and give them a few buttons
10:47 < kanzure> is the concern that the animal may not always be focused on the screen and may be aloof for several moments from time to time? is that an important issue?
10:48 < fenn> this is like chatting to a clever LLM over IRC though, not a consciousness expanding radical alteration of self
10:49 < kanzure> hm? wake me up next time you're able to do exponential self replication of silicon LLMs. i'll take my exponential biological LLMs thanks.
10:50 < fenn> we already have exponentially self replicating biological LLMs
10:51 < fenn> the problem is bandwidth and specificity
10:51 < fenn> i have to sit here and think for a good ten seconds how to formulate my thoughts as a serial stream of ascii characters, then waste a bunch of time mashing buttons with meat appendages
10:51 < fenn> that thought was already there to begin with, the transformation to serial format is a bottleneck
10:54 < fenn> it does help to have a fixed representation of a thought to go back and remember, but this could be done better and broader with a short term memory neural prosthetic
10:54 < hprmbridge> soul_syrup> But why not input the signals directly? It wouldn't need to be high resolution
10:55 < fenn> like i am re-reading the backlog while thinking about what i have already said, in order to plan out future words
10:56 < fenn> soul_syrup the visual processing region is only a tiny part of the brain, somehow your activations have to get all the way over to the other side of the brain, and information is lost in the journey
10:57 < hprmbridge> soul_syrup> The output could be formed at condensed points
10:57 < fenn> maybe a closed loop feedback system where you are monitoring activations to optimize and input signal that perfectly matches the internal representations
10:57 < fenn> an*
10:58 < fenn> what's a condensed point?
10:58 < hprmbridge> soul_syrup> The reason for direct sensory input to the brain was to control the environment so that the stimulation process and such have less noise and more effectiveness
10:58 < kanzure> fenn: i think that someone should study the extent to which other humanoids re-read the backlog or things they just typed. i think most people just type something out, hit send, and then never think about it again. or once they read your message they don't check it again.
10:59 < hprmbridge> soul_syrup> It would also allow more freedom of creativity with situations
11:00 < hprmbridge> soul_syrup> I think hybrid is the key
11:00 < kanzure> nobody knows because nobody has tried to increase the intelligence of biological neuronal matter in an orchestrated or lab setting
11:01 < kanzure> hybrid or not, we don't know. it may be the case that octopus neurons are sufficient and just need a little bit more selective pressure to get the octopus to form coherent images or whatever.
11:01 < fenn> hey that's a good idea
11:01 < kanzure> it may be the case that most mammalian brains are on the verge of higher cognitive abilities, or maybe they are already calculating arithmetic and have no reason or way of expressing it
11:01 < hprmbridge> soul_syrup> But since this would be a synthetic brain, we would have more influence on those factors, as well as the connection locations
11:02 < kanzure> a synthetic brain would be a cool thing to have, but for now i think you are going to be stuck with actual mammalian brains and engineering those
11:02 < fenn> or octopoid brains
11:03 < fenn> vampyropoda
11:03 < kanzure> i think octopus would be more difficult to work with because of all the sea water
11:03 < hprmbridge> soul_syrup> The information from certain parts of the brain would be condensed into specific signals like EEG in a way
11:04 < fenn> coleoidea*
11:05 < fenn> wut. "A unique trait of the group is the ability to edit their own RNA" (coleoids i.e. octopus and cuttlefish)
11:05 < hprmbridge> soul_syrup> We would be using human brain
11:05 < kanzure> human brain is more difficult to acquire and they don't usually let you work with those
11:06 < hprmbridge> soul_syrup> That's why we would grow it as organoid first
11:06 < hprmbridge> soul_syrup> Like how they implanted them into mice
11:07 < kanzure> soul_syrup: do you have any thoughts as to why the current results with organoid neural tissue cultures have been disappointing (eg just playing pong or balancing the demarse fighter jet simulation)?
11:08 < hprmbridge> soul_syrup> We could make the skull of the octopus very large
11:09 < hprmbridge> soul_syrup> Yea I'm fairly sure it's because they're just too small and not made of multiple regions
11:10 < hprmbridge> soul_syrup> It's like expecting an DNN with 5 hidden layers and 10 neurons each layer to speak with you
11:10 < fenn> https://www.theatlantic.com/science/archive/2017/04/octopuses-do-something-really-strange-to-their-genes/522024/
11:10 < hprmbridge> soul_syrup> Neural net stacks are like the artificial equivalent to brain regions
11:10 < fenn> not really
11:12 < fenn> GPT-4 runs on nodes of 8x A100 GPUs in a box, that's sort of like a cortical column. there are different versions of these models trained on different domains, but also a common core of language and general knowledge. tokens get routed to different nodes depending on which domain they best match (code, literature, science, etc)
11:12 < fenn> you could say the different nodes are like different brain regions. it's somewhat of a stretch
11:13 < fenn> a single cortical column has multiple "layers", that's a much better analogy
11:15 < fenn> there's lots of messy back and forth communication, and recurrent feedback loops in a cortical column, and this isn't present in current-day neural net architectures (it could be, but it would change the algorithmic properties like being guaranteed to finish after a certain time)
11:35 < hprmbridge> soul_syrup> the layers are one important aspect, but the human brain evolved to move in a 4d reality, whereas pong is just X and Y
11:37 < hprmbridge> soul_syrup> there are major differences between the artificial neural nets and biological, but what I was aiming at saying was that the current experiments that are published are probably going in the wrong direction in general
11:38 < hprmbridge> soul_syrup> the 4d would need more than just something like a simple neural network
11:39 < hprmbridge> soul_syrup> in terms of brain regions and neural net modules, there are uses for each, and just like LLMs and such, you cant have one type apply to everything
11:40 < hprmbridge> soul_syrup> like I'm now using one type of LLM that has certain weights and such, but it wont be able to properly control a 3d agent
11:42 < hprmbridge> soul_syrup> also, since we design artificial neural nets with specific initializations and rigid structures, which is completely different to biological nerual nets, so we would need to push the octopus brain in a way that would use different principles
12:02 < kanzure> you've lost me. are you going to build it?
12:05 < hprmbridge> soul_syrup> thats the plan
12:19 < hprmbridge> nmz787> https://sites.google.com/view/stablediffusion-with-brain/
12:20 < hprmbridge> nmz787>  https://cdn.discordapp.com/attachments/1064664282450628710/1154859848236802129/Screenshot_20230922-122022.png
12:39 < hprmbridge> soul_syrup> I think I saw this or something like it before, its so cool
12:42 < hprmbridge> soul_syrup> so we could encode as well, but it would input a hallucinatory reality to the brain, but if it's visual experiences are that sort of hallucinatory reality, then it shouldnt make much confusion
12:45 < hprmbridge> soul_syrup> I dont know that seems really complicated
13:48 < fenn> some very impressive text to speech synthesis (4-12GB VRAM or CPU) https://suno-ai.notion.site/Bark-Examples-5edae8b02a604b54a42244ba45ebc2e2 https://github.com/suno-ai/bark
14:06 < hprmbridge> soul_syrup> Oh yea we use that at Neocadia
14:06 < hprmbridge> soul_syrup> But not deployed yet
14:17 < fenn> why does notion suddenly hate me
14:18 < fenn> was playing audio samples just fine, now i can't play any embedded media (this format can't be played on this device)
14:26 < fenn> i bet bark could benefit from batched inferencing to generate multiple streams of audio in parallel, if it's memory bandwidth bound like LLms
14:29 -!- flooded [flooded@gateway/vpn/protonvpn/flood/x-43489060] has joined #hplusroadmap
14:33 -!- test__ [flooded@gateway/vpn/protonvpn/flood/x-43489060] has quit [Ping timeout: 252 seconds]
14:33 -!- test_ [flooded@gateway/vpn/protonvpn/flood/x-43489060] has quit [Ping timeout: 252 seconds]
14:35 -!- test_ [flooded@gateway/vpn/protonvpn/flood/x-43489060] has joined #hplusroadmap
14:35 -!- test__ [flooded@gateway/vpn/protonvpn/flood/x-43489060] has joined #hplusroadmap
14:36 -!- flooded [flooded@gateway/vpn/protonvpn/flood/x-43489060] has quit [Ping timeout: 260 seconds]
16:23 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has left #hplusroadmap []
17:17 -!- darsie [~darsie@84-113-55-200.cable.dynamic.surfer.at] has quit [Ping timeout: 264 seconds]
18:01 -!- L29Ah [~L29Ah@wikipedia/L29Ah] has joined #hplusroadmap
18:08 < sphertext_> Eli: it's hard to figure out what you reply to, as you are using a bridge. so pls add context to your messages. I don't know which compound the first part refers to ("what is the proper dose response..."). 
18:09 < sphertext_> Eli: re: antioxidant supplements, you should bare in mind that, even in healthy ppl, mitotic aberrations routinely lead to oncogenesis and it's probably a good idea not to impair their clearance, and supplementation of high-dose isolated antioxidants might do that. in the end, it's about risk/benefit i guess. i'm just not a fan of tweaking, with indiscriminate precision, an arbitrary point of a complex system.
18:11 < sphertext_> best to nuke the whole system and re-iterate from scratch (e.g proteomic clearance in plasma) 
18:11 < kanzure> a few users are using the discord channel and bridge.
18:11 < kanzure> but IRC was first :)
18:11 < sphertext_> IRC will be last too ;)
18:17 < fenn> an unhealthy amount of the internet has been moved into discord
18:17 < fenn> lots of interesting new AI projects i've found recently have basically no web presence, and then a ton of stuff in discord
18:19 < fenn> my understanding of the antioxidant problem is that it blocks exercise induced oxidative stress, which is a signal muscle for growth and the other benefits of exercise
18:20 < fenn> vitamin A specifically is a growth hormone, so it's a bad example
18:21 < fenn> signal for muscle growth*
18:50 < muurkha> muscle tumors are for some reason apparently not a problem?
19:23 < fenn> ah might have mixed signals there. vitamin A stimulates cellular turnover in epithelial tissues, not muscle growth
19:31 -!- yashgaroth [~ffffffff@2601:5c4:c780:6aa0:589c:394f:ddcd:f875] has quit [Quit: Leaving]
19:48 < muurkha> that's... not what you want if you have an epithelial cancer
19:49 < fenn> right
19:50 < fenn> however, "Higher vitamin A intake linked to lower skin cancer risk"
19:50 < fenn> maybe the higher turnover rate reduces the chance for mutation or metabolic dysfunction
19:57 < muurkha> interesting, could be
--- Log closed Sat Sep 23 00:00:30 2023