Be advised: this thread is not for serious in-depth discussion of weighty topics (we have a link for that), this thread is not for anything Culture War related. This thread is for Fun. You got jokes? Share 'em. You got silly questions? Ask 'em.
naraburns nihil supernum
1yr ago
(text post)
3484 thread views
Be advised: this thread is not for serious in-depth discussion of weighty topics (we have a link for that), this thread is not for anything Culture War related. This thread is for Fun. You got jokes? Share 'em. You got silly questions? Ask 'em.
Jump in the discussion.
No email address required.
Notes -
I wanted to share a kind of an amusing intersection of a bunch of topics that's often discussed here, which include some minor culture war issues. A couple of weeks ago, a software developer who goes by Vedal started streaming on Twitch using an AI-run vtuber by the name of Neuro-sama. There is one AI to learn how to play the rhythm game Osu that would play on screen, while a different LLM AI (not OpenAI or anything close to it, I think) would read chat and respond to messages using TTS as well as as subtitles on screen. I'm actually not 100% sure how the vtuber model is controlled, but it doesn't seem to be particularly active, so it might just be a generic video game NPC cycle.
Amusingly, I think the first prominent Vtuber was Kizuna Ai whose persona was that of an AI, but was obviously puppeteered directly by humans.
I actually started watching the streams about a week ago, despite not having an interest in vtubers or Osu. It's essentially a public LLM chatbot with TTS and anime girl avatar, and so the "conversations" go like you'd expect, with plenty of nonsensical statements and wild inconsistencies both between statements and within statements. I think the creator was clever to use such a child-looking vtuber model as a way to lean into this, enhancing the sense that the audience is interacting with someone who's still developing mentally, because I've undeniably found it cute and funny at times in a "from the mouth of babes" kind of way.
But of course, in a public website like Twitch, you should fully expect trolls to do troll things, and one managed to get through by getting Neuro-sama to question the Holocaust. This (presumably - I don't think there's any official confirmation) resulted in a temporary ban of the Twitch account, which is still ongoing. This sort of thing was bound to happen at some point given the current level of sophistication of LLM. And Twitch sometimes has the etiquette of the streamer reading aloud a chat message before responding to it, which Neuro-sama follows sometimes, and unlike a human, her quotation-reading voice and regular-speaking voice are identical. It occurs to me that Twitch's ban was justified given that they shouldn't particularly care that it's an AI controlling the voice, as long as the on-stream persona is still saying it. This is something anyone who wants to run an AI vtuber would have to account for, and I'm guessing Vedal had protections in place that were circumvented, and if not, certainly they must be scrambling to put one together.
It's fascinating and amusing for me to see this play out in real-time. The concept of AI-run celebrities has been around for a while in science fiction, and we're transitioning into it being just plain reality. At least, maybe; 1 example does not a trend make, and the novelty might wear off with a few iterations. Still, I'm glad to see someone experimenting with this kind of thing, and I'm surprised we're already at the point where someone is able to crudely but successfully staple together some modules to make this AI-run e-celeb. Interestingly, since the streams are on the creator's channel, a lot of the conversations are about Neuro-sama's relationship with her creator Vedal, resulting in Vedal also becoming a sort of e-celeb.
Also, recently Neuro-sama started singing full songs on stream, which is more in line with AI celebs in scifi. I don't know how that is accomplished other than "cheating" by using some preset, even if the voice is generated real-time. I'm pretty sure whatever AI model they're using isn't figuring out the pitch and pronunciation of each syllable in these full-length mainstream songs as it goes.
But yeah, fascinating intersection of AI alignment, holocaust denial, online moderation policies, and weeaboo culture that I wouldn't have seen coming even just a couple months ago.
I'm glad to see there are still true heroes out there, even on a site as awful as twitch
More options
Context Copy link
More options
Context Copy link