What are they using as input? Like, you can have software that can control a set of outputs learn what output combinations are good at producing an input.

But you gotta have an input, and looking at their products, I don’t see sensors.

I guess they have smartphone integration, and that’s got sensors, so if they can figure out a way to get useful data on what’s arousing somehow from that, that’d work.

googles

techcrunch.com/…/lovense-chatgpt-pleasure-compani…

Launched in beta in the company’s remote control app, the “Advanced Lovense ChatGPT Pleasure Companion” invites you to indulge in juicy and erotic stories that the Companion creates based on your selected topic. Lovers of spicy fan fiction never had it this good, is all I’m saying. Once you’ve picked your topics, the Companion will even voice the story and control your Lovense toy while reading it to you. Probably not entirely what those 1990s marketers had in mind when they coined the word “multimedia,” but we’ll roll with it.

Riding off into the sunset in a galaxy far, far away? It’s got you (un)covered. A sultry Wild West drama featuring six muppets and a tap-dancing octopus? No problem, partner. Finally want to dip into that all-out orgy fantasy you have where you turn into a gingerbread man, and you’re leaning into the crisply baked gingerbread village? Probably . . . we didn’t try. But that’s part of the fun with generative AI: If you can think it, you can experience it.

Of course, all of this is a way for Lovense to sell more of its remote controllable toys. “The higher the intensity of the story, the stronger and faster the toy’s reaction will be,” the company promises.

Hmm.

Okay, so the erotica text generation stuff is legitimately machine learning, but that’s not directly linked to their stuff.

Ditto for LLM-based speech synth, if that’s what they’re doing to generate the voice.

It looks like they’ve got some sort of text classifier to estimate the intensity, how erotic something a given passage in the text is, then they just scale up the intensity of the device their software is controlling based on it.

The bit about trying to quantify emotional content of text isn’t new – sentiment analysis is a thing – but I assume that they’re using some existing system to do that, that they aren’t able themselves to train the system further based on how people react to their specific system.

I’m guessing that this is gluing together existing systems that have used machine learning, rather than themselves doing learning.