home.social

#sttng — Public Fediverse posts

Live and recent posts from across the Fediverse tagged #sttng, aggregated by home.social.

  1. @philbetts @pip We will never be able to put the genie back into the bottle. The open weights models are out there, and fine-tuned versions of them with the censorship removed can be found all over the Internet. Any sufficiently nerdy person with a beefy GPU and enough RAM can just download them. 4chan /gif/ is full of sometimes incredibly silly porn versions of scenes from all kinds of visual media, but a lot of Star Trek and Star Wars, usually locally generated with one of the latest Minimax models. You can find scenes where the first half a second is just a couple of frames of #STTNG and then all of a sudden, Troi takes her top off, and suddenly Riker steps towards her with his erect penis out, and she starts to suck it. Things like that just happen now. Things like that won't ever vanish again as long as there are people with halfway decent GPUs in their PCs. Like gamers or 3D designers.

    And there are LLMs, AI agents, multimodal models, etc., running on local hardware as well, available for download, and people have been working on removing or outsmarting the built-in censorship as well. Nobody is going to stop it. "AI" probably won't spontaneously come to life anytime soon, but AI tools that can do a lot of things automatically (even if rather badly) that only a couple of specialists could do a few years ago means that we will see a lot of weird AI enhanced shenanigans all over the planet in the future. Reality as we knew it before generative AI went mainstream is over, and things will be getting very strange and surreal. And everything that can be turned into porn will be turned to porn by somebody. Uncensored fine-tuned derivatives of and LoRAs for Stable Diffusion 1.5, SDXL, Flux, Flux 2, Z-Image Turbo, and Krea 2, have already made #Rule34 explode beyond measure by making the automatic generation of erotic images of popular fictional characters possible. As soon as there is some new movie or TV show, whether live action or animated, people are training LoRAs for the characters for all kinds of video and still image generators. Sheer horniness can drive geeks to do a lot of voluntary work for the AI gooner community. LoRAs too naughty for any regular web platform can be found on torrent sites and anonymous file hosting services.

    And of course there are also uncensored LLMs fine-tuned for erotic roleplay. There are recipes for building prompts that turn such an uncensored LLM into an unlimited AI girlfriend/boyfriend/sex slave/dom(me) or whatever. If you have tens of gigabytes to spare on your hard drive, you can just install them, although they'll be very slow and far from realtime if you haven't got enough RAM and a sufficiently powerful GPU.

    As long as people with powerful PCs exist, nothing of that is every going away. Horny nerds and robot fucker wannabes will make sure of that. The technology isn't going to vanish ever again, not as long as people can run it at home even without any cloud services whatsoever.

    So everyone who wants to have access will have access. The modified models that have had their safety guides removed will persist because of the horny crowd and the wannabe hacker script kiddies. And eventually, someone will build some kind of AI worm that installs copies of itself on every machine it can hack; even though LLMs can't really think, can't really come up with completely new ideas, they can follow recipes they have learned and exploit common security flaws that have been in the training data. Somebody might even prompt the initial instance of such a worm to train new versions of itself so it can always copy what human hackers have been doing recently.

    In the world of Shadowrun, the old Internet is destroyed by the Virus of 2029. Maybe in this real world, it already happens in 2027 or '28, depending on when somebody succeeds in building a self-learning fully autonomous AI worm. I wouldn't be surprised if somebody did that. It doesn't even need to be "really" intelligent, it doesn't need any kind of consciousness or personality, it just needs to borrow from human intelligence. LLMs can't produce any radically new patterns, but they can produce all kinds of variations of already existing patterns. A friend of mine just got ChatGPT to tell a really funny joke about some apocryphal religious scripture. Another friend of mine is building a taxonomy app for Android all by himself using Claude. I wouldn't want to debug that app, but it appears to be no worse than many supposedly professional ones. As long as the problem you're trying to solve follows patterns already present in the training data, you can get generative AI to synthesize a halfway decent solution. And that also includes how to hack into a system, upload a copy of itself, and start another instance over there with a prompt that tells it to do the same to other systems.

  2. @guyjantic

    Yep, ST:TNG. He designed the #BTTF Delorean, the ST:TNG Enterprise D, the Enterprise C (but they kitbashed something instead of using his awesome custom design), and he even completed the design for the Enterprise-REFIT / Enterprise-A for ST: The Motion Picture (it was halfway done before him).

    Really clever guy. Also, if it helps score more nerd points, the reason I emailed him is that Nausicaä's Mehve in Nausicaä of the Valley of the Wind reminded me of the Enterprise D — specifically, the jet intake-thingy reminded me of the Enterprise D navigational deflector. He said it was one of his favorite animes, but it wasn't related to the D, but rather the shape of the deflector had something to do with mirroring the profile of the saucer section, instead, IIRC.

    I might be able to dig up the emails if your ex goes spastic over it. 😄

    #StarTrek #STTNG #TNG #Nausicaa #Nausicaä #anime