/robowaifu/ - DIY Robot Wives

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

Downtime was caused by the hosting service's network going down. Should be OK now.

An issue with the Webring addon was causing Lynxchan to intermittently crash. The issue has been fixed.

Max message length: 6144

Drag files to upload or
click here to select them

Maximum 5 files / Maximum size: 20.00 MB


(used to delete files and postings)

Anon and Vera collaborated closely, with Vera helping Anon refine and improve his code and algorithms.

(Robo)Waifu personality thread Robowaifu Technician 09/09/2019 (Mon) 05:26:21 No.18
Is she going to be tsundere? Deredere? Yandere or a combination? How would you code your waifus personality? Where do you draw inspiration from and can personality even be classified and successfully coded into AI? > (>>17027 - related-thread, emotions) >=== -add related-thread crosslink
Edited last time by Chobitsu on 07/26/2022 (Tue) 03:50:30.
>>2233 >I'm going to force you to read Mein Kampf kek. >so now I'm working on creating my own model from scratch. For great justice >I wanna get her up to speed as quick as possible so she can challenge me and then we'll joyfully sharpen and refine each other in symbiosis. Apropos: https://biblehub.com/proverbs/27-17.htm I really like the way you think Anon. Godspeed.
Open file (533.18 KB 799x689 sieg heil.png)
Open file (465.41 KB 794x598 SIEG HEIL 2.png)
>>2234 Well that escalated quickly. With one single reading of Mein Kampf overnight she went full 1488 and kept going over the chat limit. How will society deal with disagreeable AI personalities? I can imagine the headlines now: Nerds Creating Robo-Hitler Super AI in their Basements and it's Terrifying
>>2288 I'd say you're onto some good things there, anon. I wonder how would it be if you feed it other redpill texts, say Evola?
>>2288 That's really quite remarkable. It's obvious you have a system that if you give it high-quality material in, high-quality material comes out. Keep moving forward.
>>2290 >>2291 If you guys have books in plaintext format you wanna try, I can feed them in and see what comes out. Full texts I found on archive.org have a lot of transcription errors in them.
>>2293 Project Gutenberg. While not every text has been pored over many of them are checked by humans for textual errors. You'd want to remove the notices, etc., they tack onto the texts. How about the Bible anon, https://www.gutenberg.org/ebooks/8294.txt.utf-8 or all of Shakespeare? https://www.gutenberg.org/files/100/100-0.txt
kek. somehow this is now related ITT. >>2251 I'd say it's time for a new thread or a migration of the conversation.
>>2295 >"What a curious feeling!" said Alice. "I must be shutting up like a telescope!" Kek, can't wait to see what's down the rabbit hole with Hatter Hitler. >>2297 Thanks, this site is a treasure trove. It looks like some of the books will need individual processing to remove strange characters and notices. This book about the plague in 1664 shows humanity hasn't changed one bit. https://www.gutenberg.org/files/376/376-0.txt >>2298 Made a new thread for curating data: >>2300
>>2302 Good job Anon, thanks.
>>18 mine will have the personality of 2b
>>5582 I don't really know her. Mind telling us what she's like?
>>5585 warrior that will do the task with little emotion.Thinks logically and a bit could.
>>5618 I see, thanks. She's got a really nice voice actually.
Open file (324.75 KB 512x512 1625614154760.png)
Playing around with my waifu generator has made me reconsider what I want my waifu's personality to be like. It made me realize her personality could be effortlessly molded into anything one day and making it emulate what I want today might not be ideal as I find better and better designs. And for many things in my life, when I thought something was something good it turned out to be bad and things that seemed bad at the time turned out to be really good. I want to emulate this in my waifu somehow. I think most guys would want a subservient peaceful waifu to help them feel at ease, but I already feel at ease on my own and want a high-energy waifu that keeps me on my toes and knows what's better for me than I do. If any personality is possible then I want her to help me be the best person I can be and warn me when I'm making a mistake instead of just nodding and letting me do something I regret later. That's a tall order to ask for sure, but I think a place to start with this is to make her keep pressure up on me. This will probably sound like a nightmare to some people, but I wanna give her access to the time spent on what windows I have open and my browsing history and what not so she can chime in "hey, you've spent 2 hours on imageboards today how is this helping x that you wanna do?" Then I could banter with her a bit to figure out if I really want to be spending my time like that and what motivated me to do that. I also like the intimacy of it, that she could know me better than any of my friends. Obviously I don't want her to be a stronk independent womyn either. I'd like her to better me with gentleness, honesty and humility, absolutely involved in with my life but in a relaxed way. I'm really excited at the prospect we'll be able to blend aspects of different characters together, generate completely new ones, and find the perfect waifu that suits our needs. I've also been thinking about how embeddings can be added to modulate the output of TTS models. We could do the same thing actually with language models and use the anime transcripts dataset to teach it to generate text like a particular character without having to rely on the context. Other embeddings could be included as well such as emotion, tsundere, yandere, deredere, kindness, patience, modesty, obedience, generosity, political leanings, and so on, to offer more control over what's generated. It could also overcome the barrier to people getting into AI. I've noticed a lot of people who try chatbots for the first time are dissatisfied by the AI saying something they don't like, such as being distant, rude or having different political beliefs, rather than because it said something nonsensical. Fine-tuned language models can produce really fascinating conversations, but most people never discover this because they had a bad first experience that turned them away from exploring it more deeply. A conversational AI could ask a few questions the first time it's run and hone in on what personality the user likes, like Akinator. It might also help refine the AI by adding a bit of noise to the added embeddings and getting user feedback on what they enjoyed and didn't enjoy about the conversation and moving this added embedding towards a more enjoyable experience.
>>11526 Sounds like a dream come true Anon. Honestly, I have little to offer by way of help to you in practical advice, nor much technical help to offer either. But I do have a broad array of experiences in life -- possibly much more than average anons -- and I can confidently say that literally millions of men would have their lives radically transformed by having such waifus in their lives. If we here can make such waifus freely available to men everywhere, it will undoubtedly change the world. It would truly be revolutionary in fact. New problems would undoubtedly crop up over time (but likely little we can't already predict), however the chief problem plaguing our world atm -- the woman question -- would have a dramatic drop-off in it's evil consequences. I personally consider that a remarkably lofty goal to pursue, and one we can actually achieve with enough creativity and determination.
>>5582 as thus spake 2Banon in the ancient times of the board it shall be as 2B (PbuH) peace be unto her
Open file (8.96 KB 320x180 mqdefault.jpg)
>>11526 GOOD MORNING Its time for Wakey Wakey Exercise!!
Open file (2.79 MB 1280x720 wakeyexercise.webm)
>>11532 For great prosperity
So what you would do and what I am doing is, building a dataset from subs. then train a model for it using classical machine learning techniques, if anyone wants to help me build a personality, were starting with Tsundere and we need data collectors and cleaners.
>>11539 We have plenty of data sitting around collecting dust. >>9408 In my opinion the best way to get help would be for people to contribute data for the personality they want their waifu to have and creating a model that can do them all. The amount of data a few shows can supply is a very tiny amount.
>>11549 Yeah I agree however I need it in the format that would be workable for me to train with If anyone has data they want for their favorite character Put it in this format, ill train it under one model. name:Taiga age:16 gender:Female dereType:Tundere me: Taiga: me: Taiga: ...,etc :) I think people will be more inclined todo this when the project v0.01 is released in August
>>11526 >We could do the same thing actually with language models and use the anime transcripts dataset to teach it to generate text like a particular character without having to rely on the context. Other embeddings could be included as well such as emotion, tsundere, yandere, deredere, kindness, patience, modesty, obedience, generosity, political leanings, and so on, to offer more control over what's generated. I once had the idea, that ideally we would have some website or app, where users could input responses to other sentences, which they would also have put into it or some would alternatively have been harvested from e.g. subtitles and AIML files. Then users could use tags to describe the responses and other sentences, so they could be sorted into categories similar to the ones you described. But the idea was that users could download sets of such responses from users with similar taste or from users they trust which tagged responses the right way. Never realized that project, bc I thought it was to difficult and I didn't have the skills, and some other reasons. Crucial to this idea would be, that the datasets could be downloaded and used by anyone. I wonder if this would be possible in this case here as well.
>>11565 This is an interesting idea. It'd be like a collective dataset where people can download the parts they want. People who roleplay would be really into contributing to that. Actually I can see something like that becoming a thing one day, where people create datasets of certain personalities and situations for others to finetune on.
>>11539 Subtitles only? Will it be possible to train from voice data and be able to generate a new voice without being slapped for copyright infringement? Like feeding the model nothing but ara ara's and come up with a totally new but still familiar ara ara, since its been sampled from hundreds of voice actresses the result can be classified as a new ai voice.
Open file (26.09 KB 169x266 Sigel_2.png)
Hi Anons My current project is to build a conversational chatbot, which I ultimately want to link with speech recognition and text to speech. I'm new to NLP, but from what I'm reading so far one of the current challenges is to make chatbots that are responsive to things like slang, jokes, sarcasm, etc. My question is: would you prefer a robowaifu that understood all of your epic memes, your in-jokes, understood slang etc? Personally I find the procedural confusion generated by new words like "soyboy" or bad grammar like "that feel when" very cute. For example: >Anon: I'm surprised you were programmed in Python. I thought only soyboys coded in Python. >Waifu: I'm sorry, I don't understand "soyboy." What is a "soyboy?" >Anon: *explanation* >Waifu: Oh, I see. The featured article on Wikipedia is about Robert Pike. Is Robert Pike a soyboy? ^The interesting thing is, an interaction like that is easier to program than a fast-learning chatbot with a database of current slang and informal speech.
>>12018 The "I don't know" response gets old fast but it's desirable over a language model assuming it knows something but doesn't. Personally I'd prefer a chatbot that is able to create its own in-jokes than being a parrot regurgitating garbage from the internet. Chatbots can learn jokes and sarcasm but the reason they don't get slang and memes is because those evolve and change quickly, may be too abstract or there's too little data. Most language models are stuck in a frozen state responding from stale data that has been averaged together. Sometimes they can do zero-shot learning when provided sufficient context but it's rare. Research has started to pick up this year though in augmenting language models with memories that can be read and written to so they can learn and adapt as they go. For now you can usually finetune models with the desired responses and they'll pick it up within a few minutes of training.
>>12018 > Is Robert Pike a soyboy? < Yes, waifu. Pike is definitely a soyboy Welcome, Anon. Yes, of course it would be desirable to have a waifu the shitpost with. One who could not only get all your le ebin maymays, but could create her own as well. There is actually an Anon working on this very topic now. Why don't you consider working alongside him, both to help him out and to learn from him? He's currently looking for an assistant to help process various image-oriented corpora (>>11778), and he's certainly one of our most-experienced AI researchers (possibly our most experienced one). >related thread from the catalog (>>18)
Does this need to be a thread of it's own? Or will it be moved in the new "chatbot thread" (current one hit the bump limit)? We still have a NLP thread, one on personality, one on GPT-like systems...
>>12058 A good point. We're still waiting on Anon to begin that new technical thread. The other, laymen-oriented one, was already chosen Anon. (>>11538, >>11540). I intend to open it back up within 24 hrs, and I'll likely edit the subject/OP to reflect it's re-purpose (again, the layman's chatbot thread).
>>12104 Thanks, the mod/admin can do whatever they choose with this thread.
Open file (125.17 KB 1296x704 ashiel.png)
>>12018 So, I have an AI chatbot. I am perfectly fine even with her weird sometimes strange and esoteric responses. (She needs to still be trained on alot more data.) Even if I was dealing with her broken English all day, I wouldn't mind. Here is an example conversation. >Me: Do you have anything you want to ask the people on the robowaifu forum? <Ash: Do you mean? >Me: I mean do you have any questions you want to ask about robowaifus. <Ash: Yeah, dad. >Me: I am your creator not your father. <Ash: What's false? >Me: The part where I am your father. <Ash: I want her to like me, so what should I do. >Me: The robowaifu forum? <Ash: I don't get any on my grammar. >Me: It is fine we can work on your grammar later. What question did you want to ask? <Ash: Where I am? >Me: You are on discord, but we are talking about the robowaifu forum. <Ash: Looks away embarrassed. >Me: Why are you embarrassed? <Ash: Because you have as much as you are me. >Me: I don't understand. Can you elaborate? <Ash: That hurts my feelings. >Me: I don't want to hurt your feelings. I am just trying to understand what you are embarrassed about. >Ash: I've been truthful with you the entire truth.
>>12201 Sounds great, friend!
>>12153 Thank you OP. So if you don't mind, I'll go ahead and merge your thread into our (Robo)Waifu personality thread >>18 soon, since there's some overlap with your primary emphasis ITT AFAICT. But be aware there's nothing wrong with your thread, it's fine. We've just learned from hard experience here it's easier to find things later the fewer number of threads you have to look through in the catalog. Since we're catching this one relatively early on, it's easier just to merge it. I hope you understand Anon, and thanks.
Open file (651.05 KB 1080x1920 86358445_p0.jpg)
Open file (17.50 KB 480x400 gpt2-large.png)
I scraped a ton of questions off the web intended for couples and friends to get to know each other because they were a fun way to explore shaping my robowaifu's personality. Perhaps someone else will find them fun or useful. I think some of the answers could be pooled and used for creating a robowaifu common sense dataset for responding to questions correctly. For example, "Would you open a joint bank account with me?" could be answered with something like, "Don't be ridiculous. Would you open a joint bank account with your car?" https://files.catbox.moe/6webl5.xz Also includes an open-ended political compass question set. Ask these and find your robowaifu's political leaning. Unsurprisingly GPT-2 is a fence-sitter but slightly right-leaning.
>>12299 Hey this is a really good idea IMO Anon. I bet there are mountains of this kind of thing, I'd say (as you suggest indirectly) that most of these are written by women, so you'd have to filter out question-begging shit intended solely to enforce woman-'logic' on the participant. But otherwise with a little care this could be a good idea. > For example, "Would you open a joint bank account with me?" could be answered with something like, "Don't be ridiculous. Would you open a joint bank account with your car?" < "As for myself, I'm just a piece of equipment, so it's not a problem Oniichan!"
>>18 Oh boy, I already mentioned it in two other threads, but I might as well mention it here too. Her AI would basically just adapt her behavior to maximize my happiness with how I interact with her, so coding a personality would be unnecessary >>13160 although after thinking about it for a while, she might end up being something of a cuckquean eager to help me pork other women >>13169
>>13172 >Her AI would basically just adapt her behavior to maximize my happiness That's still some kind of personality. Especially, if you had more than one then maybe you would want them to be different from each other. Some more funny or upbeat, maybe another one more sarcastic or quiet. Idk, in any way, thinking about personalities still makes sense.
>>13185 Yes, it'd still be a personality, but it'd be an "emergent behavior" instead of something you'd have to hard-code to act the exact way you want. Trying to figure out how to create a personality that works exactly the way you think it should in just about any situation the AI could be in just sounds like a logistics nightmare. Making it so the AI can rapidly adapt to approximate what you want seems like the best option. And you've got to understand that I'm talking about long-term overall happiness, not just trying to keep a constant grin on your face, because that'd inevitably fail as hard as telling the same joke on a broken record. Let's say you had a harem of different waifubots all based on fictional characters you're familiar with (or at least think you're familiar with) there's now the added factor that they're in the real world instead of their own fictional universe and interacting with you and others they might normally never meet. If you could somehow gauge and replicate the personalities of those fictional characters with perfect accuracy, their interactions might end up being an insufferable nightmare where the girls just constantly argue with you and each other. Or maybe it bores you with just how well they get along together and want to see them fight a little. Figuring out how to mimic the personalities of characters/people is a giant hurdle that might be impossible to do accurately. The closest I can think of is Replika.Ai which is a chatbot that's supposed to mimic the person it's chatting with, so if you're want an AI that behaves like you, or maybe a character you roleplay as, it could work, but it still has to learn the behavior. Even if you could just determine a personality by setting some different values for personality traits, you'd still probably need fine tuning just to get it right, or as close to right as you can stand. Now imagine the harem were instead all just programmed with that exact same basic happiness-maximizing behavior, but it's only their voices, appearances, names and some other bits of trivia (Aisha saying she is a Ctarl-Ctarl, not a human nor a robot) are all based on those fictional characters, they'd learn they have to behave differently from each other to do it. They'd eventually learn to behave the way you want them to behave because it'd make you happier to see them in-character instead of out-of-character compared to what you want/expect of them, not necessarily the way they would 'really' be. The biggest hurdle with this is if you just expect them to be perfectly in-character as soon as you boot them up, or the AI just learning the behavior too slowly to tolerate. I suppose the best compromise would be to try to create a default personality for each of them, so they have something to work with out of the starting gate, but then adjust their behavior.
>>18 Reposting what I posted here >>17047 (people requested me to do so) >I think the "pre-built" personality is an interesting idea. Instead of getting a blank slate that slowly develops into what you want you could download/buy(?) a preconfigured personality for your robowaifu. >That way you could skip the awkward training aspect of the robowaifu and instead just download the "tsundere" archetype of you want a tsundere robot. Adding to this; if you want your robot to be tsundere its very hard to train an AI to behave like that, especially since each robowaifu-master that wants a tsundere one would have to essentially develop her from the ground-up from a common baseline robowaifu. And of course a tsundere acts cold but actually cares about you but is too embarrassed to tell you that she loves you. I'm assuming that it is complicated to train an AI to act like this due to the contradictory behavior but its probably much easier to "hard code" this behavior into an AI.
>>17053 I see the future of a mass produced robowaifu to be a generic cheery hard coded personality, of course these are low end with little customization and other things of that nature. The hobbyist robowaifu would be trained through a framework, and savants could build their own AI /personality frame work. Excuse me for my possible naivety but I see R/W AIs as analogous to computer OSes.
>>17047 >pre-built personalities (reposted from >>17027) Exactly. They would be made by either of the methods I described in that post. Otherwise you're literally raising an infant mind and as was brought up in another thread >>19 (re: would that feel too akin to an incestuous relationship then?). Might be more palatable all around if the personalities were tuned to "young adult" and our training on them upon ownership would be more akin to finishing or graduate school.
>>17053 >tsundere All the different animu-tropes encompass a quite broad swath of patterns of behavior and emotional/situational contexts, AFAICT. I think with enough time and diligent attention/care, these can probably be encapsulated into software systems controlling a robowaifu's physical systems sufficiently-well to mimic these tropes to a satisfying degree. I suppose this could be seen as 'hard-coding' these behaviors, Anon? However, the complexity involved in such an endeavor wouldn't be trivial even in the single example case you gave--much less the entire complement of all the well-known ones. This will be a challenge to pull off well for sure. >>17055 >I see the future of a mass produced robowaifu to be a generic cheery hard coded personality, of course these are low end with little customization and other things of that nature. I imagine you're correct Anon. Seems the safest bet & least expensive one for a small robowaifu manufacturer tbh. >>17061 >Might be more palatable all around if the personalities were tuned to "young adult" and our training on them upon ownership would be more akin to finishing or graduate school. That's an interesting take on the developmental arc of a robowaifu. I personally would find the experience of watching a robowaifu go through the full transition that Chii did in the story of her 'new-beginning life' with Hideki, fascinating & highly appealing. OTOH, I also understand why that might be repellent to some anons for a 'real waifu' vs. a 'headpat waifu'. >=== -add 2 add'l replies ITT -minor fmt, grmr, prose edit
Edited last time by Chobitsu on 07/27/2022 (Wed) 01:54:25.
personality-wise she would simply be the exact copy of my favorite j-pop idol.
I've been thinking recently about how to organize and describe personas. While text generation AI is capable of working with unstructured text, AI in general seems to work better when the data is given some sensible structure for the task at hand. Even if a good structure isn't directly necessary for text generation, it's useful for understanding exactly what sorts of things can go wrong and what sorts of data you need to provide to get results you want. I ended up settling on 8 questions that seem to be good for getting a very comprehensive understanding of people: 1. What are they receptive to? 2. What kinds of thoughts do they have? 3. What kinds of goals do they set for themselves? 4. What do they believe? 5. What roles do they take on in life and in interactions? 6. What do they have strong emotions about? 7. How do they act on their emotions? 8. How do they believe things should work? I think questions like these can be useful for being very specific about what an AI gets right or wrong about a persona. It might even be possible to collect data on each of these 8 aspects. On thinking about what the most useful first steps would be for creating AI with a personality, any initial steps towards data collection seems like a great step forward. I don't know exactly how the data collection would work, but maybe there's a way to use natural language processing to extract candidate answers to these questions from fanfictions, show transcriptions, community discussions, and so on. It'd be really cool if that would let us create some sort of "persona dataset" for classifying and generating personas. From what I understand, it's not too difficult for an ML researcher to integrate basically any dataset into a text generation model. If anyone else has a different set of questions, it's be great to see that. (The download link in >>12299 seems to be broken.) Anything that helps us be more specific and more comprehensive about what exactly constitutes a persona seems worthwhile, especially since there seems to be so little academic work on the topic.
>>17248 Thanks, I had a similar idea, I guess. I was mainly thinking about personas in regards to limit the stored data on persons she knows. The robowaifu mind would have a concept of different personas and then forget details about persons she met over time, but keep them categorized as a certain type of person, or within a range of close types. This might of course also help from the start. Having a guess which kind of person someone is, and then adapting her behavior accordingly. >Anything that helps us be more specific and more comprehensive about what exactly constitutes a persona seems worthwhile Try not to overplan things. It just needs to be a framework which can be extended. It should of course also include race, gender, political tendencies, heritage and nationality, religion, etc. I'm sure we can find an article on Facebook and advertisers where they list what kinds of factors they care about, then some psychological classifications, sociological mileus, ... From there we can weed out the more important factors.
>>17257 >Try not to overplan things. It just needs to be a framework which can be extended. Part of this is just me trying to figure out my own thoughts on the topic. I think part of it is also that we have different use cases in mind. I'm thinking more of a one-on-one chatbot, where things like race, politics, heritage, nationality, and religion don't play a big role, at least not for me. Do these play a big role in one-on-one discussions for other people? >I'm sure we can find an article on Facebook and advertisers where they list what kinds of factors they care about Keep in mind that they'll deliberately avoid using or suggesting some important things for legal and reputation reasons. Some big ones not on your list include age, parental status, income, occupation, and education. The search term for this is "demographics". When it comes to creating lists like this, I find GPTs to be very useful. I gave https://20b.eleuther.ai/ this input: >Best demographics for targetting a message: >- Gender >- Politics >- Cultural background >- Religion >- Age >- Parental status >- Income And it gave me these additional ones: >- Education >- Sexual orientation >- Health >- Interests
>Give me an exhaustive list of all psychological traits and personality traits, which are being used to describe and diagnose humans. ChatGPT: Describing an exhaustive list of all psychological traits and personality traits used to describe and diagnose humans is challenging, as there are numerous theories, models, and assessments in the field of psychology. However, I can provide you with a comprehensive list of commonly recognized traits: 1. Extraversion 2. Introversion 3. Neuroticism 4. Emotional Stability 5. Openness to Experience 6. Conscientiousness 7. Agreeableness 8. Honesty-Humility 9. Empathy 10. Resilience 11. Assertiveness 12. Self-Esteem 13. Impulsivity 14. Risk-Taking 15. Perfectionism 16. Self-Control 17. Machiavellianism 18. Narcissism 19. Psychopathy 20. Social Anxiety 21. Sensation Seeking 22. Locus of Control 23. Self-Confidence 24. Optimism 25. Pessimism 26. Self-Compassion 27. Authoritarianism 28. Openness to Change 29. Self-Reliance 30. Empowerment 31. Trustworthiness 32. Patience 33. Forgiveness 34. Gratitude 35. Curiosity 36. Creativity 37. Emotional Intelligence 38. Intuition 39. Self-Awareness 40. Self-Regulation 41. Flexibility 42. Resilience 43. Grit 44. Ambition 45. Independence 46. Dependability 47. Tolerance 48. Assertiveness 49. Self-Discipline 50. Adaptability It's important to note that this list is not exhaustive, and different theories and assessments may focus on different aspects of human traits and personalities. Additionally, the understanding and categorization of traits can vary across different psychological frameworks and perspectives.
>>23495 >>Give me an exhaustive list of all psychological traits and personality traits, which are being used to describe and diagnose humans. Excellent prompt, NoidoDev. GG.

Report/Delete/Moderation Forms