/robowaifu/ - DIY Robot Wives

Advancing robotics to a point where anime catgrill meidos in tiny miniskirts are a reality.

The canary has FINALLY been updated. -robi

Server software upgrades done, should hopefully keep the feds away. -robi

LynxChan 2.8 update this weekend. I will update all the extensions in the relevant repos as well.

The mail server for Alogs was down for the past few months. If you want to reach out, you can now use admin at this domain.

Max message length: 6144

Drag files to upload or
click here to select them

Maximum 5 files / Maximum size: 20.00 MB

More

(used to delete files and postings)


Knowing more than 100% of what we knew the moment before! Go beyond! Plus! Ultra!


Open file (410.75 KB 1122x745 birbs_over_water.png)
Open file (99.96 KB 768x512 k0p9tx.jpg)
/robowaifu/meta-5: It's Good To Be Alive Robowaifu Technician 03/07/2022 (Mon) 00:23:10 No.15434
/meta, offtopic, & QTDDTOT General /robowaifu/ team survey: (>>15486) Note: Latest version of /robowaifu/ JSON archives available is v220523 May 2022 https://files.catbox.moe/gt5q12.7z If you use Waifusearch, just extract this into your 'all_jsons' directory for the program, then quit (q) and restart. Mini-FAQ >A few hand-picked posts on various topics -Why is keeping mass (weight) low so important? (>>4313) -HOW TO SOLVE IT (>>4143) -/robowaifu/ 's systems-engineering goals, brief synopsis (>>16376) >--- -Library thread (good for locating topics/terms): (>>7143) >note: there's also a searching tool coded right here for /robowaifu/ that provides crosslinks straight to posts on the board. it's named Waifusearch, and the link to the software is provided inside the Library thread's OP. -Latest version of Waifusearch v0.2a (>>8678) -Latest version of BUMP v0.2g (>>14866) >--- There's a design document for the specification of general design and engineering choices for our basic Robowaifu Reference Model A Series (TBD). Please have a look at it, and collaborate together with us on it ITT Anon. -Robowaifu Design Document (>>3001) >--- -previous /meta's: (>>38) (>>3108) (>>8492) (>>12974) >=== -add team survey notice -edit thread headliner
Edited last time by Chobitsu on 07/21/2022 (Thu) 21:41:47.
>>16936 The easy solution to the space problem is chopstick legs. Like an impossibly thin magical girl. If she only needs them for her Anons use then, they don't need to support her.
>>16937 You somehow misunderstood. The skirt would be out of plastic with wheels, there's enough space for the legs, but the skirt is much wider than the body. I'll most likely add an opening to the side and cut them off that way. Using the skirt for stabilization is one of the reasons to have it in the first place, of course. But she doesn't need more on the sides than in the front and back. The poles will become bows connected to the sides which are left over, which will also look much better.
>>16938 I did indeed presume that the space for her lower body was small. What are the dimensions for your current design? MaidCom is currently being designed with a hip 150mm wide and 78mm deep. Legs will likely add to the 150mm width when plugged in. I do like where your design thoughts are heading.
>>16939 The design is completely flexible. It's just a sketch for now, but also in OpenSCAD, so it can be changed by changing some variables. I'm actually not formally part of the MaidCom team, yet. I never got to register to the board. >hip 150mm wide and 78mm deep Huii, that's small. Good to know.
>>16940 You are formally invited to join. I'm looking forward to us creating cute girls
>>16941 Thanks. I look into it, must read the thread for registering.
Open file (38.56 KB 425x612 Skirt_T01v01b-001.png)
Open file (37.58 KB 425x612 Skirt_T01v01b-002.png)
>>16938 Played around with rotate_extrude() to create some better connectors for the stands. This is too annoying right now. I'm probably going to make a thread for the waifu mobile skirt or whatever it will be called at some point and finish it. Or I'll go on in the prototype thread with it.
Open file (167.16 KB 1280x1707 Untitled.jpg)
>>16943 A dedicated thread for mobility skirts would make sense for keeping the concept easily accessible for Anons.
Please pardon the interruption in this off-topic tred :^) but this is absolutely amazing tbh!
Open file (25.85 KB 474x700 2blove.jpg)
Hello /robowaifu/! It's been quite a while since I posted on the board, Probably more than a year if I had to guess. I think the last time I seriously posted was when Chobitsu went MIA and everyone thought he was dead. Wow, that was a lifetime ago but it only feels like it happened yesterday. So, /robowaifu/, how are you doing?
>>16951 Well, he isn't dead. There are at least two projects going on to get closer to a actual robowaifu build. One very small rolling one and one more like a animated but not very mobile lovedoll. The guy working on his ML based AI also seems to be on it, and there was some development around Robowaifu Foundations (C++ Software). In between there was a bit of drama, also more people started to register to the board somehow and organized as a team. Nevertheless it seems to have slowed down and there isn't much activity. Maybe because it's summer, and the one after Covid (or is it)?
>>16951 > So, /robowaifu/, how are you doing? Pretty alright, there is some success. The largest problem will be a large-scale training run with volunteers of course. I will test various scenarios on my infrastructure, but in the real world various unexpected errors happen. Still, looks like training a model with 1-2 billions of parameters is doable, which is, for current generalist agents, state of the art. The upper limit for the current architecture is likely 10B. The most interesting part will be optimizing the NN architecture to achieve maximum efficiency at our parameter counts, and deciding on the final composition of the multimodal dataset. >guy working on his ML based AI also seems to be on it It is true. If someone is willing to help me, I could use volunteers to start testing our distributed computing framework.
>>16953 Could you please respond to my question on whether or not a OnePlus 8 would be an ideal mobile computer for MaidCom? I'm very close to getting one for her but, I'd like to know your thoughts on the matter. For the distributed network, I have a Ryzen 5 3500u machine with 12GB's of RAM and a 256GB SSD I could dedicate to the purpose for a few months. I can install any OS on it though, I would prefer a Linux distro if possible.
>>16951 Welcome back Anon! :^) >>16952 >Maybe because it's summer Wait isn't Eternal Summer supposed to be more activity? :^) JK, yes you're probably right. I myself hope to get more done more regularly once the year starts again in September.
Open file (2.39 MB 2160x1904 pinkhairARK3.png)
>>16951 2B is Love 2B is Life (PBuH) Can I ask if you're the original guy who posted pic related (see spoiler)?
>>16948 I love it The stars are our destiny, I've had to deal with the absolute smooth brains on SM who are, as grown adults, convinced that thousands of scientists, NASA Employees and billions of dollars are all wrapped up in a conspiracy to make us believe in outer space (when I suppose we live in a giant snow globe I guess?). It's exhausting and one of the 3 reasons I just quit all my SM except a few Discord servers. My biggest regret in this life is that I'll probably die well before realistic robowaifus and space arks are a thing. Maybe someone could save some DNA and clone us in the future? FYI I think it's very small brained to assume that belief in "space" or not should preclude a belief in a higher transcendent power. IMO an incomprehensibly massive universe should only open up the possibility even more.
>>16977 >smooth brains on SM >I just quit all my SM except a few Discord servers. Ahhh, you mean social media. I needed two minutes or so till I realized. >someone could save some DNA and clone us in the future This would be your twin, not you. Consider having children. Either way, we might go through a filter first and no one knows how this is going to turn out.
Open file (286.16 KB 816x500 smug_hubble_is_smug.png)
Open file (31.53 KB 500x500 DTTC_audio_cover.jpg)
>>16977 >IMO an incomprehensibly massive universe should only open up the possibility even more. Just so. Einstein himself was """morally""" offended by the metaphysical implications of his own equations of General Relativity regarding the historical nature of the universe. Pic-related was the moment of revelation for old Al, that yes, the universe is expanding. > [1] >t. currently devouring pic-related > [2] BTW, if you'd like to go deeper, I'm all ears. But perhaps we should move this to the Christianity bread instead? (>>2050) 1. https://digital.archives.caltech.edu/islandora/object/image%3A2013 2. https://support.reasons.org/category/featured-products
>>16961 >whether or not a OnePlus 8 would be an ideal mobile computer for MaidCom Certainly it is not ideal, but it should be workable with an optimized NN, and I will develop said optimizations for a snapdragon 855 I have. Still, I don't think it's worth it to buy the device ASAP, a cheap used device will serve you just fine. The SoCs are improving, so it's obviously better to buy a latter once it hits the secondary markets. >>16977 >I just quit all my SM except a few Discord servers. You could hint me at some of these so I could join them. Or not. >Maybe someone could save some DNA and clone us in the future? Yes, you can do it by storing your tissue and/or by sequencing your genome and storing it on a durable medium at various locations of the world. Still, if it's possible, it's much better to have a lot of children with a person you like.
>>16961 >I have a Ryzen 5 3500u machine with 12GB's of RAM and a 256GB SSD I could dedicate to the purpose for a few months. I can install any OS on it though, I would prefer a Linux distro if possible. Thanks for being open about it! We can use this machine to help us generate the dataset (specifically, expert playdata from a pretrained RL agent over a specific game), I will ping you when I'll have a client ready to run. Linux is alright with me.
>>16994 discord.gg/nFdr5KT9 SoaringMoon's discord, I'm a mod there I also have a server but I'm only using it to post notes and collect links but I was thinking of opening it up if there was an interest discord.gg/fkkamSCQ Also twitter, but be warned I follow some lewd accounts so I am not responsible for what you might see My twitter @RoninMeta Had run into Kiwi on FB and sorry to deactivate after such a short run, but tbh even the R/W groups I was in or created were only for the feels or lewds and not generating interest otherwise. I decided the entire app/site was not worth it Fb had become too much negativity and cliquish drama. I think I was just using it as a case study of the degradation of interpersonal relations, civility and male-female relations. Point is received and now a hermit I shall remain.
>>16984 >>16994 I do have a child, one who is now of legal adult age. I feel like I've mentioned it already in here but I am Divorced: details are dull, tbf it might have been better to have not even married but I did it for my kid even if I did not make it the full 18 years So yea I am coming at this from a slightly different angle, but if I can be perfectly honest it (robotics, AI, human-AI relations, and space exploration) these topics are simply the most interesting thing to me right now. This short little spit of a lifespan we're given, I want nothing else other than to have a shot at pushing the forward edge of "what is possible" and what we are capable of becoming.
>>17003 The ideal thing to do is to use Twitter and Reddit to only follow people talking about tech and cutting out everything else, while staying rather anonymous. There are many data scientists on Twitter and there's a lot of hobby/maker/tech related groups without politics or drama on Reddit. Then there's also Quora, Hackernews and Stackoverflow. But even then, there's the risk of getting distracted by getting into more and more topics one is interested in, without doing much while just watching others or commenting a little bit. It's better just to use it if one really has a question, but other wise staying away from it. (I wish I would follow this rule myself, lol)
>>17005 I have one or two Reddit accounts. I only lurk there because it's a normiesphere of mainstream groupthink but actually I think it was a dead reddit group that led me here
Open file (1.07 MB 1571x1555 space_setller_compass.jpg)
Open file (5.34 MB 2499x3900 07-solar-sail.jpg)
Open file (133.80 KB 804x1280 ringworld_02.jpg)
>>17003 Thx, joined both. Not a big fan of discords, but it is what it is, and we are who we are. Maybe in the future we'll make a leet IRC or a private chat. >>17004 Feel for you, anon. Still, my highest respect for bringing a child into this world, imperfect as it is. It's really hard to procreate these days due to known reasons. Surely your descendant will grow into someone bright, who will appreciate the tangle of history he is going to be a part of. >but if I can be perfectly honest it (robotics, AI, human-AI relations, and space exploration) these topics are simply the most interesting thing to me right now. >I want nothing else other than to have a shot at pushing the forward edge of "what is possible" and what we are capable of becoming. Can relate, and I have a hard-won set of beliefs and life decisions related to this outlook. We are not the only ones and not the most well-resourced people who ponder these questions since becoming barely adult-minded. One can see this attitude in a few silicon valley higher-ups, and in several of the C-levels heading the AI companies engaging in scaling race right now. In an age where you can fold a protein on a single GPU with a subsentient language model doing most of the work. Regardless, I think there is a place in the future for us, if we make use of the opportunities floating nearby now and then. >>17005 Agree, though as a decade+-old HN user it's pretty obvious that these are mostly diffused info sources and nothing beats a focused project you own. There is always more generalist expertise one can learn from such media, but the chances of the metaphorical stars aligning in such a way as to replace a long focused deliberation/prototyping/engineering run are close to 0.
>>17003 Gorgeous 2B, Meta Ronin. >>17004 >I want nothing else other than to have a shot at pushing the forward edge of "what is possible" and what we are capable of becoming. I hope for far, far more for you in the future Anon (indeed for everyone here tbh). Certainly we can still achieve much in this life but it will take a metric boatload of hard work! :^)
Open file (674.39 KB 700x3285 list of pizzas.jpg)
Open file (535.07 KB 1000x909 yorha 2b royalty.jpeg)
Open file (148.58 KB 1020x1600 yorha 2b ass. jpeg.jpeg)
>>16952 I'm glad to hear that. I made a mistake and took roughly a 2 year break from college follow the start of corona chan's party. I do recall a separate board being created that had something to do with rule 34 but if it's board drama, I don't personally care for it. >>16953 Who exactly are you? Back in the day when i used /robowaifu/, no one named fagged. I take you're a significant programmer or engineer? >Pretty alright, there is some success. The largest problem will be a large-scale training run with volunteers of course. I will test various scenarios on my infrastructure, but in the real world various unexpected errors happen. >Still, looks like training a model with 1-2 billions of parameters is doable, which is, for current generalist agents, state of the art. The upper limit for the current architecture is likely 10B. >The most interesting part will be optimizing the NN architecture to achieve maximum efficiency at our parameter counts, and deciding on the final composition of the multimodal dataset. Pretty much everything you said went over my head. I guess I got more to learn, don't I? >>16976 No, that isn't me but I do remember that thread. I wonder how that guy is doing. Who are you, by the way? I see you as "meta ronin" but what do you do here specifically? Are you a vol or just someone who namefags?. Also does your name have any meaning? Were you once a samurai? If so, what happened to your lord? >>16974 >Welcome back Anon! :^) Thanks chobitsu. I think the last time we spoke you were still going to college or was it university? Have you finished yet? As I mentioned previously, I am still going to college. /robowaifu/, I have some more questions for you all. I'm a pretty frequent poster of /monster/. Recently, there have been some pretty questionable posts. For one, shareblue, a democrat "activist" group that kept arguing in favor of abortion posted there again until they were banned. Also there have been a significant amount of glow posts on the webring in general. Mainly links to "questionable" sites. Has this kind of activity been happening here? What pizza would /robowaifu/ eat from image 1 related?
Open file (102.09 KB 1280x720 powerofcute.jpg)
>>17004 >(robotics, AI, human-AI relations, and space exploration) these topics are simply the most interesting thing to me right now. I decided to be more interested in longevity than space exploration. Longevity is partially more similar to developing robowaifus, since individuals can do something by themselves. It's also much less pie in the sky. Fantasies of space exploration and the way these stories are told is something that was very likely pushed by media to support the idea of a unified mankind. Also, making stuff and preparing for problems (prepping) is more relevant, imo. Robowaifus could be the "perfect preppers bride". Space exploration becomes more relevant after we went through the next filter.
>>17131 Pizza bait is getting reported by anyone seeing it and then it's getting deleted. Political discussions are being moved to the bunker or deleted, dependent on their relevance to /robowaifu. We don't need to agree on everything outside of the scope of the project at hand.
>>17132 >Robowaifus could be the "perfect preppers bride" This is at least partially true for many anons, I expect. But, you can't really expect to survive the "Good times right around the corner!" phase of the full cycle if you don't get /out/ & /fit/. Also, power generation is definitely going to be a hot topic for robowaifu masters, along with spare parts & other maintenance items. Once we have them going, it will be good to go waay /out/ for several shakedown dry-runs, yeah?
>>17131 > I do recall a separate board being created that had something to do with rule 34 /clang/ is linked in the welcome tred (>>3), as well as /monster/. They both have robobutt materials. >Back in the day when i used /robowaifu/, no one named fagged Lol. Sorry, but I made it pretty clear that namefagging was OK here basically all the way back at the beginning of the board on 8ch. No anon has to do so, ofc. As time has progressed however, most of us regular anons recognized the value of it, since we're basically a creative group engineering endeavor together (as discussed here >>10469). It's working well for us, actually. >Have you finished yet? As I mentioned previously, I am still going to college. Not yet, I'm about halfway now, slow going. Good luck with your studies, Anon. >What pizza would /robowaifu/ eat from image 1 related? I love a good deep dish. But I'm going to have to spoiler your images, b/c >implying implications You can freely-discuss DFC-related issues here (>>2666). Also, <1m high robowaifu designs are actually of great engineering value--particularly during these early prototyping years (square-cube law & whatnot). But as you're well-aware, it's an issue fraught with threats for any man b/c feminism & 'muh anti-pedo hysteria'. What robowaifus you want to create on your own are entirely your own business ofc, but as a board we need to keep things under a tight reign here to protect the site itself, as well as Robi. >tl;dr This isn't /b/, Anon. Don't push it.
>>17131 everything you need to know is here >>15854 >re: namefagging I considered that only a faux-pas on /pol/ or /b/. I think for the purposes of this project, keeping track of who is posting what and learning each others worldviews and idiosyncrasies has been a net asset. If anyone really cares to dox based on what I post here knock yourselves out, I've always been into robotics and AI as both speculative science fiction and something I want to see become a reality in my lifetime. Seeing as how every sexual deviation under the sun is now tolerated and celebrated it would be rather hypocritical of those same people pearl clutching at the prospect of somebody getting physical with their companion AI bot. (though no doubt some will, and already are) As far as the greater mission, I'm here to learn, lend advice, brainstorm/spitball ideas and help build up the robowaifu community in the most beneficial and globohomo-free way as possible. (Think EFF but applied to waifus) I'll be hosting within the next month or two at robowaifu.link and robowaifu.online both volunteering web space for anyone's related website and as a linktree or webring of sorts for robowaifu resources. I've already posted links to a couple discords (I know how much some of you hate them, but so far it is working really well for a few of us) my own and another anon's. If you're still part of chan culture and into monstergirls I don't take any issue with this, but the tone here has been set by Chobitsu and that's one of the reasons I stay and participate but I'll still lurk around lainchan, /pol/ and a few others
Open file (274.19 KB 888x500 2Bfall.jpg)
Are we due/overdue for a meta-6 as it is a month away from being Fall already?
Is it time for an official art thread on /robowaifu/? There does not seem to be one from what I could find searching through the archives.
>>17387 I would say to go for it. It can't be any worse than half the shit found on twitter and elsewhere.
>>17392 We need to develop the visual robowaifu concepts further, using novel degrees of freedom provided to us by brand-new freely available tools like Stable Diffusion.
>>17275 That's just a little rhyming quirk I did for my small attempt at humor; the /meta 's aren't actually seasonal. The only req. I have since /meta-2 is that they intentionally be off-topic for the OP. >>17387 >Is it time for an official art thread on /robowaifu/? Past due time. >>17401 >Is it the maidcom one? Yes, but we're all working on whatever we see fit as well.
>>17408 >Obsession with DA JOOZ >Also your quick reply box doesn't work. Go back. This is not /pol/
>>17407 >the /meta 's aren't actually seasonal. that lent a comfy aesthetic to the board though
Open file (37.97 KB 480x360 hqdefault.jpg)
This feels like the best place to post this, and this seem necessary to post. The subject line is about the people on this board (though no personal message is intended) rather than any of their potential creations. Currently, this board is following approaches which will make it impossible for its robowaifu technology to be strategically surprising to the forces which adapt everything to loop around to nowhere, except the direction which advantages them. It doesn't make sense for this post to cause a sudden change in approach, leaving past projects here (which are often personal projects not just /robowaifu/ projects) to waste (though this is a sunk cost fallacy). However, /robowaifu/ should observe its own approach and examine its possible effects of instead of simply riding the wave. The Robotics Approach: Distributed Processing doesn't make sense. Splitting functions by locality makes even less sense unless the scale is thousands of kilometers. Even the human body largely doesn't do this, despite nerve signals travelling far slower than lightspeed. Reliability is better served by duplication. Sane design given technology demands that there will be software running on compute hardware that exists to run that software as well as possible given constraints, and that there will be robotics hardware controlled by the software. There will be two distinct sides to the problem, with distinct solutions. This is obviously exemplified by the idea of a robowaifu with compute external to the body. This is of course assuming no philosophical issues which would make everything done here a total loss. Robotics-only approach: sex doll Software-only approach: visual waifu/chatbot In each of these domains individually, /robowaifu/ has not produced anything surprising or state of the art. Could the combination of these domains provide an advantage? Robotics->Software: Embodied cognition. Given the right (novel) software approach, perhaps this could make the AI more interesting. Avoiding the way of current large language models, which simply continue text without being in any world, and only have the last few thousand words and trained structure as context. The software-only alternative is simulated environments and recurrence. Software->Robotics: Obviously forget about any safety standards (when applicable) here. Software could do more with less complex mechanisms, potentially making some of the robotics side easier. These are only the technical advantages of course, the combination itself has an advantage as a product even if there is no advantage to design the parts from the whole. Product: At what point does the combination make a product superior to either a sex doll or chatbot individually? I don't see the combination being designed from the start. It is better for software to have uniform or easily specified hardware, and it is better for robotics to not be locked to any software. The natural conclusion is that development of robotics and software should be separated except for when developments in the other can be utilized to provide an advantage. In other words: Design an open source robotic doll kit, which is easy to program. Many people develop software for this kit. Once software advancement reaches the point that better hardware is needed, release v2 doll. Repeat. Or: Design an AI model with very multimodal and general capabilities. People build robotic hardware as desired that can be controlled by these models. Software improvement occurs simultaneously. Repeat. Is faster than: Build hardware and software for a robowaifu v1. Improve hardware and software until v2 is needed. Build entirely new hardware and incompatible software for v2. Repeat. It is also more consistent with free software (and hardware) principles and collaborative method to separate robotics and software. AGI or just superior to females? There are approximately two different kinds of people on /robowaifu/: Those whose first goal is AGI and a waifu (waifu/): The question of AGI has not been discussed in sufficient detail here. A lot of the work here also seems not to have realized the failure of symbolic AI. What intelligence means has also not been investigated. This side favors the software approach, unless philosophical issues change the problem. Those whose first goal is ending the female advantage, and want to use the advantage of combining robotics and software to achieve this (/robo): The question of strategy needs answering. The question of artificial wombs too. A theory of society and sex is needed for effective action. This side favors the hardware approach, but wants as much software as is necessary to gain enough advantage to capture demand. Some questions: What happens if the system embraces the idea of robowaifu, and use their resources to make a version superior to what we could build and provide it to the masses? But this version of course has a catch. /robo says: This would never happen, they aren't going to give away their power irreversibly like that for no reason, robowaifus are a red line they must supress. But what if the catch is no artificial wombs? What if they use permitting robowaifus as a political pressure release valve, while taking artificial (and female) wombs under total political control. Their win is even more guaranteed then, just delayed a generation. What happens if they just copy and paste any AI development /robowaifu/ makes, and just use their superior resources to adapt it for their goals? waifu/ says: The capabilities /robowaifu/ wants to develop are not useful to the system, there is a qualitative distinction within AI development, like how 3d printing and GPS advantage different sides. Is this distinction real? If we believe the normal theory, that intelligence is optimization, if we believe in orthogonalism and universalism, then no. All intelligence is the same and liquid, and unless you get a head start on recursive self-improvement their win is very likely. If you want to actually take a qualitatively different approach (which I believe is necessary for real AGI) then this must be
>>17419 (post was truncated) If you want to actually take a qualitatively different approach (which I believe is necessary for real AGI) then this must be addressed first. Might continue..
Primary goal poll: https://strawpoll.com/polls/6QnM7YN3AZe >>17419 >What happens if they just copy and paste any AI development /robowaifu/ makes, and just use their superior resources to adapt it for their goals? >waifu/ says: The capabilities /robowaifu/ wants to develop are not useful to the system, there is a qualitative distinction within AI development, like how 3d printing and GPS advantage different sides. I would answer this differently. If the system embraces the idea of robowaifu, I would say that's great. That's a tremendous amount of resources and intelligence it's going to swing around at accomplishing my goals. I just want waifus, and I don't care to have a monopoly on them. It might even be better for me if we didn't have a monopoly on them because then the bar for societal acceptance would be lower. >Design an open source robotic doll kit, which is easy to program. Many people develop software for this kit. >Design an AI model with very multimodal and general capabilities. People build robotic hardware as desired that can be controlled by these models. >Could the combination of these domains provide an advantage? I think you're trying to work out what are the easiest paths to interesting outcomes given decentralized development. Uncoupling things lets each track move faster, moving faster seems necessary when we're trying to "catch up" with all of the capabilities in relevant fields, and catching up with the capabilities in relevant fields is very useful for creating interesting domain combinations. It's difficult for a decentralized group to do monolithic projects because it's difficult to coordinate work and resources. It's easier to take the monolithic results of centralized groups and modify/connect/build on them. That makes me think, for development work specifically, maybe it would make more sense to create threads around projects-we-want-to-build-on, rather than creating threads based on knowledge-we-want-to-consolidate or outcomes-we-want-to-accomplish. So for example, we'd have a Stable Diffusion thread focused on manipulating the SD model, or we'd have a GPT thread focused on using GPT-2/GPT-J/... in interesting ways, and we'd have a KoboldAI thread focused on using KoboldAI. That would mean specifically keeping the GPT and KoboldAI threads separate, even though they both try to accomplish text generation, since bootstrapping people for building on GPT is very different from bootstrapping them for building on KoboldAI.
>>17421 These are excellent points. Mind you anyone coming here under the assumption that any of us have command over a production line or an actual physical company with capital is going to be sorely disappointed. While these things are respectable goals to work *toward*, I think most of us are just brainstorming how we can leverage what's available for us to scrounge, reverse engineer, 3D print, etc. No doubt at least one person in here has access to a foundry and a CNC machine but it's all too much work for one person or even few people, no matter how sufficiently motivated. I've seen good things with VR/AR and that might have to be the bridge of "make do" until someone tosses us a rope of an generic customizable humanoid robot (which IMO to do it right will take some retooling of common mechanical parts and electronics, think: how human prosthetics are designed versus industrial robots)
>>17421 >If the system embraces the idea of robowaifu, I would say that's great. The question intended for waifu/ was what happens if they use AI advancement partially driven by /robowaifu/ to improve their uses for AI: mass surveillance, social control, etc. Is a future where robowaifus are possible but where the technocrats also have all the tools needed to end history survivable? I doubt it. AI development alone doesn't change the power balance towards us. Before, they were few in number, with limited attention, and had to remain hidden to not be overwhelmed by everyone else. Now they have A100s, and you do not. Changing this means finding some technological asymmetry, which means going beyond the standard view of orthogonalism and optimization. Cryptocurrency is one example of a very asymmetric technology, but AI currently is not (it may be asymmetric against us in fact). >>17423 >While these things are respectable goals to work *toward*, I think most of us are just brainstorming how we can leverage what's available for us to scrounge, reverse engineer, 3D print, etc. Of course right now it doesn't really matter. But in general before deciding how to build, better decide what to build, and why.
>>17419 I'll be merging this thread into our /meta thread soon, just to let you know OP.
>>17423 I understand. On the AI side, it doesn't take a large amount of resources to build on Stable Diffusion or GPT-J 6B. You can train new tokens for these models ("softprompts", "textual inversion") on Google Colab for free, and you can rent an A100 to fine-tune models for, currently, $0.35/hour, or an RTX 3090 for $0.04/hour. In some cases, it's feasible to train massive models from scratch for free because there are groups (Google TRC, Stability AI, CoreWeave) that give free compute resources for open source AI research projects. I'm not familiar with robotics, but my understanding is that a lot of the design work can be done using free software (MuJoCo, SketchUp, IsaacGym). The actual manufacuring and real-world tests might get expensive, and I don't know what it would take to create models for off-the-shelf parts (would manufacturers be willing to provide it?), but maybe those things can be worked out later. I'd find it very exciting if we could get a good waifu design & simulation working with mostly-plausible components and environments, even if we didn't know how to manufacture it. >>17424 >The question intended for waifu/ was what happens if they use AI advancement partially driven by /robowaifu/ to improve their uses for AI: mass surveillance, social control, etc. In that case, I would say that's an acceptable risk because a world without waifus is unacceptable. I would take steps to prevent people from misusing my [hypothetical] advances for surveillance & control especially if that surveillance & control inhibits waifu progress, but I would not try to prevent misuses if it means stopping waifu progress. I'm curious how others feel about this point. https://strawpoll.com/polls/Q0Zp4kle6ZM
>>17430 a world with total surveillance is almost unavoidable except if someone lobbies for "privacy zones" where drones are restricted, or build a faraday caged home, etc. No one outlawed security cameras and while it undermines privacy such measures should reduce petty crime and dangerous crime in urban areas perhaps making them safe again one day? tl;dr total privacy is a lost cause, better learn to deal with it. Those in power who have a lot to lose if their private stuff gets out will surely put in some limitations that the rest of us can benefit from too. That's how the legal system works
>>17430 also saying, by the time waifus with govt. or corporate backdoors might be a standard thing, it may be a moot point when there are "fly" drones at a density of a few per some odd cubic meters of any public space already
>>17415 Duly noted Meta Ronin. I'll see if I can come up with something sufficiently cheesy for /meta-6. :^)
>>17430 >and you can rent an A100 to fine-tune models for, currently, $0.35/hour, or an RTX 3090 for $0.04/hour Please share where you can get such cheap compute. Lambda is $1.10/hr for an A100 and A100s and 3090s are never available on vast.ai.

Report/Delete/Moderation Forms
Delete
Report