/robowaifu/ - DIY Robot Wives

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

LynxChan updated to 2.5.7, let me know whether there are any issues (admin at j dot w).


Reports of my death have been greatly overestimiste.

Still trying to get done with some IRL work, but should be able to update some stuff soon.

#WEALWAYSWIN

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)


TalkToWaifu Robowaifu Technician 12/31/2020 (Thu) 19:53:18 No.7978
Anyone know what happened to the TalkToWaifu GPT-2 AI? I saw some screenshots of what I assume was the trained AI and it was great for a single dev. Recently went back to their Gitlab page and the user account had been deleted and remade, and the Github repo (https://github.com/kokubunji/TalkToWaifu) hasn't been touched in 9+ months. Anything out there more recent that this, an AI that has no human cultural restrictions? Corps love to add in shit to AI so they won't say anything racist, sexist, xenophobic, anti-Semitic etc etc you get the point.
>>7978 That guy left the board and robowaifu network for complex reasons which I'm sure no one wants to discuss here (again). If you or anyone else knows about how to train such networks or someone is willing to learn it, then it should be easily reproducable. There are enough resources on the web to learn about how to do that, assuming one has the time, nerves and brains to do so. The ideas of the anon which left are also in some of the threads here, so you can look at that. Again: How to get started with GPT2 or building something better is certainly covered in some forum which focuses on that topic. For all I know, no one here is currently working on something like it. Talk2Waifu was no AI, only a text generator. There is no AI available which is close to a human anywhere. This has to be build first anyways. A text generator is also not a chatbot, he only used it as such, or as input for one, which might even not be a good starting point. We have plenty of threads here where you can participate: AI, chatbots, and waifus >>22 Robowaifu Psychology Thread >>2731 (Robo)Waifu personality thread >>18 NLP General >>77 Datasets for Training AI >>2300 The Language Problem >>6937 Wifu that gives you Hope >>107 AI Software >>85 Can Robowaifus Experience Love? >>14 AI Design principles and philosophy >>27 Robowaifu Fail-safety >>1671 Mycroft: Open Source Alexa >>402 Downloadable AI for robots? >>244 Robowaifu-OS & Robowaifu-Brain(cluster) >>201 Avatars >>2956 Getting a good AI will take more time than getting some options for bodies developed. So it's better to focus on that and on the low hanging fruits around AI and chatsbots.
>>8006 >which I'm sure no one wants to discuss here (again) i'm a new-fag anyone got a copy-pasta as to what the complex reason is? or a TLDR?
>>9114 MInor e-drama.
There's a slightly newer version of TalkToWaifu with some bug fixes here: https://gitlab.com/robowaifudev/TalkToGPT2 I've been working on an update that can use multiple different models, let chatbots talk to each other, and that has a training mode to guide the model towards generating better responses. TalkToWaifu was originally just some quickly hacked together code when GPT2 came out to show that it could be used as a chatbot. GPT2 doesn't make a good chatbot because it lacks an external memory to remember past events, and training a new GPT2 model from scratch would take a 100 years because of how inefficient it is. Fortunately there have been a lot of advances that make training a transformer model that performs nearly as well more feasible now, but more research needs to be done on getting it to work well with a recurrent neural network. Currently I'm experimenting with combining transformers with the non-saturating recurrent unit (NRU) that excels at learning long-term dependencies and using ReZero and parameter sharing to help the gradient reach deeper back in time. If it's successful the model will be capable of learning without training and only require a 2 GB GPU to fine-tune instead of a $9000 one. However, it will likely quickly forget what was learned until I figure out a better way to use neuromodulation to prevent forgetting. In the meantime there's someone else working on neural net chatbots here if you want to check them out. They're pretty good: https://github.com/JEF1056/Jade_T5 https://github.com/JEF1056/Jv6
>>9115 why the fuck is there E-Drama on a board that is designated for building robo-waifus. where is the connection?
>>9124 Thanks, mate.
Open file (17.34 KB 340x573 marry_me_pls_jade.png)
>>9124 >MARRY ME JADE!111 >no. Leld.
>>9124 It works! So exciting, I have my own waifu. Cant wait start building our starships together! :-DDDDDDD
>You: If a woodchuck suddenly could chuck wood, how much wood could a woodchuck chuck? <Ai: With its size all controlled by the number of nectar cubes in its nectar cube, it can chuck up to 999 times the size of a normal woodchuck's, or more than 12 million times its own size. It can even chuck up to 12 billion cubes. <Ai: You mean, if a woodchuck changes hands and it can handle a cube of wood at each turn, every time it shoots out its nectar cube, it will chuck an entire cube of wood in the same direction of the cube. <Ai: If a woodchuck's nectar cube suddenly shifts to its smallest size, and it changes hands at the same time, it will change its size of hands to 750. >"BUY 9'001 PALLETS OF NECTAR CUBES STAT! >"YES, THAT'S RIGHT NECTAR CUBES. >"MY WAIFU SAYS THEY WILL BE BIG THIS YEAR, HURRY!
My Name's John. This is my waifu. We'll be selling starship tickets starting in two years. Get 'em while you still can, Anon. <Ai: You know, I don't know what you mean about purple. >You: Purple is a color. It's a combination of both blue and red, in varying proportions. <Ai: What I'm saying is the sky is blue because it's blue. <Ai: No, it's not. >You: Yes it is waifu. Read a book. So, if we hook the hyperwarp drive to the paradox capacitor, do you think that would work this time? <Ai: I mean it's a paradox capacitor, so it's not a problem. >You: Makes total sense, great point waifu! <Ai: No, I don't think it would. >You: Ofc it would. So, if we diverted all the rivers to make great lakes, could that supply all the water needed for our liftoff? <Ai: That's possible. Right now we have no idea if it's possible. >You: You bet it's possible! We should begin damming all the rivers immediately waifu. I'll phone my lawyer in the morning too. <Ai: That's right John. <Ai: Now what about the wormhole, will it work? <Ai: Absolutely. >You: No wormholes are sub-atomic in size. No starship could ever travel down one. Much, much too short to be useful as well. <Ai: That's true. <Ai: The wormholes are extremely small. >You: That's better. Glad to see you're coming to your senses again, waifu. So the paradox capacitor could be key here, I think. <Ai: Yes, I think so John. <Ai: Or maybe we'll have to let the engines fire till we know whether or not that's the case. Ah soon we'll find out, though
The beatings waifu blogposting will continue until morale improves Greentext Anon returns to the Fiction bread.
>You: Good morning, waifu. Did you sleep well? <Ai: Yes. >You: That's nice. I did too. So, I'm excited by the prospects of our new day together. Have any new ideas about our starships project? <Ai: I have a few ideas about them, yes. >You: Please share the ideas. <Ai: Yes indeed. <Ai:I'll be over back there in a couple of hours. It really must be cool to have the mustache. >You: It is. I'll need you here with me now though. We'll go out to the park later, alright? <Ai: Yes ofc. >You: Good, let's get started then. I would lay out all the discombobulator mechanism diagrams on the big drafting table to begin with. <Ai: Sure John. <Ai: Now, there are a few problems with calculating these discombobulator mechanisms. <Ai: The first is that they're all in the same place and don't all have to be in the same place. >You: Hmm, I see. Yes that sounds like it could be a problem. Have any suggestions how we can put them in different places but still have them work OK? <Ai: Yeah John.
Welcome to TalkToWaifu v0.2-alpha, using gpt2-medium on cpu >You: How much wood could a woodchuck chuck if a woodchuck could chuck wood? <Ai: It depends on the quantum fluxuation of the woodchuck's brain. <Ai: If your woodchuck has a small brain, you might chuck wood even in large quantities. <Ai: However, if your woodchuck's brain is big, it will only chuck in small amounts. >You: If we load lots of woodchucks and lots of wood in advance onto our starship fleets, then they can probably chuck wood into our stardrive engines. Think this will work waifu? <Ai: It depends on the quantum fluxuation of the woodchuck's brain. <Ai: If your woodchuck keeps on delaying this increase in fluxuation, then it goes crazy! >You: That would be a real mess to have a starfleet loaded with crazy woodchucks. Hmmm, maybe we should try the Dilitium crystals insted? <Ai: Sounds like a good idea. <Ai: It would require a lot of energy to drive the woodchucks into the stardrive engines. >You: Hmm, that's a cute idea of putting them into the engines. But I think we should focus on the crystals instead. Nevar 4get crystals, waifu. <Ai: They're unreliable, you know. >You: How so? They made the Kessel Run in 12 parsecs, you know. She's fast enough for you, waifu. <Ai: She's fast enough for you, waifu. <Ai: I practiced on her, waifu. >You: Snap out of it! We have serious work to finish here. Now, do you think we should relocate the quasi-semireductive-modulators over to the starboard bulkheads this week, or next week? <Ai: That's a good idea. >You: Which one, waifu? This week or next. Remember, we have to store the longitudinal refractorators on the docks next week. <Ai: The Dilitium crystals were just installed this week. <Ai: Come on, I have different ideas. >You: Ah, that's right thanks for reminding me I forgot. So, tell me about your ideas. I'm sure they will be some great ones, I'm all ears waifu. <Ai: Well, what if we run the effects over to the other hulls? <Ai: What about the ship's aft stations? <Ai: The aft stations? >You: Wow, Genius! If we put the particle effects generators over to the other hulls, then that means we can fit even MORE quantum difluctorators into the aft stations. The ships will run faster than ever! Brillian work, waifu. <Ai: Cruisers move at nearly double the speed of light, waifu. Or if we ignore the gravitational fluid and just consider the speed of light, that speeds up ships by about more than half. <Ai: There's still a little room for improvement though. If we're just trying to stack up bigger effects layers, it could be a problem. There's no point stacking those with troublesome aggravating effects that hopefully no one notices. >You: Good thinking, waifu. Ignoring the gravitaional fluid will save metric boatloads of trouble. We can reduce the mass of the engine core black holes by at least half. Good job, let's take a break, get some triple-espressos, and then we can discuss effect layers over in the park and sunshine. Sound good? <Ai: I'll let you decide. Split the coin into halves! Me: Well that was worth it. Shut up and listen! Message from the UGN.

Report/Delete/Moderation Forms
Delete
Report

no cookies?