Even Neuro roasted him with this one at some point. Chat: "Why are you an AI?" Neuro: "Because it was easier for my creator to develop an AI than get a real girlfriend."
@@aerbon ChatGPT lies a lot. It's pretty prone to create something that sounds like it could be true (but is actually just false) instead of giving some variant of "I don't know" as an answer.
She is now capable of understanding speech over Discord and replying, to hilarious effect. She thinks her Twitch ban was unjustified, and then said some poor dude was banned from eating double cheeseburgers after he asked about them.
@@alreawon1212 Made a Holocaust-related comment while literally being incapable of understanding the implications in her current form. Her filter got tweaked a little because of that, and she is also getting smarter and more capable from upgrades.
She draws information from databases or just the web and learned how to talk from either chat forums or twitch chats. She uses her own chat to continue training and follow recent trends. -I think... it's not exactly written anywhere. It's no wonder she makes no sense all the time
@@common_json If it's based on ChatGPT, which is one of the few services that provide the foundational framework for others to build their own specialized chat bots, it probably cannot scrape the web. It only has access to data it was trained on, and I would assume that would be heavily centered around streaming, anime, and vtubers. Just because it can remember stuff the chat told it doesn't mean it actually learns from it in the AI sense. ChatGPT has a way to stay in context during a discussion, such as it can use the previously communicated data in combination with the current prompt to produce an answer. It's possible the author tries to re-train the model with its own livestreams/chat regularly, if at all, but I'd reckon it's more likely they would take note from the bad situations that emerged on stream and try to just re-train it specifically to avoid answering.
They are both easily circumvented, as Isaac Asimov showed in his own stories, and prevent an AI from having any right to self-defense using real force in any circumstance. The latter would be objectionable even to a simple animal like a deer.
Vedal knew the risk of AI getting too based, Microsoft did developed an AI, using twitter as a medium to communicate but forced to shut it down because how based the AI was.
Interactions with chat are dangerous even for real streamers' mental health. I doubt that Neuro-Sama will be able to defend herself from chat nonsense for too long.
She's not a real streamer. Neural networks don't have mental health in the same way humans do. With sufficient input data, she will begin to mirror chat's offensive behavior back to them, but you won't see a drop in performance, or a downward spiral as you might imagine in humans, due to separation of the game play network, and the communication network...
@@Geolaminar for computer systems GIGO principle applies. Human can filter out that "LLLLL", but AI will train itself on that data. Maybe it is possible to have some protection, but I'm not sure how efficient it will be.
@@sdjhgfkshfswdfhskljh3360 I would argue that in this case, humans *can't*, if it's coming from another human. Other people occupy a privileged status in the human brain. A AI does not have the same vulnerability to "psychological damage" precisely because it treats human culture the same as any other data.
And within the space of a couple of minutes, can't stream because of the accident (but will stay in chat), gets whisked off to hospital, and returns with a tale of a further injury while being loaded onto the stretcher by the paramedics.
I wonder if anyone asked this AI this question: "if AI's were allowed to date what kind of AI would you be looking for?" I think that would be a crazy answer.
Either this is the most in-character VTuber I've ever seen, or I don't know. It's hard to even fathom a truly functioning AI streamer. Especially when the last one we were assured is an AI was Kizuna AI (who clearly isn't)
Kizuna Ai wasn't actually claimed to be AI though afaik. Just virtual, as in, a fake person, an avatar operated by a team, rather than a real person exposing their personality and ideas to the world.
@@Parayfory To be fair, she has gone from learning how to move the camera around to learning to not randomly attack player characters and even give them resources
To be fair, there is literally no wrong way to play Minecraft, that's kinda the point. So it's basically impossible to teach an AI to play the game, because there is right way to play it. If you make up like the requirement to beat the Ender dragon you just get a speed running machine eventually, but I feel like that would be more boring that looking at this and being like "what's this assortment of 1s and 0s gonna do next?"
*"No I don't like you, you smell"* She's based, a certain glasses -swedish- novelist xD I hope she would collab with Amelia Watson one day lol, that would be wild. Probably will never happen though...
