Forum - View topicINTEREST: Eiichiro Oda Asks ChatGPT to Write the Next One Piece Chapter
|
Goto page Previous 1, 2 Note: this is the discussion thread for this article |
| Author | Message | |||
|---|---|---|---|---|
Davy Sprocket
 Posts: 24 |
|
|||
|
There's an AI "anime" channel on Twitch. It's funny, but I miss the Seinfeld one.
AI does seem like the future though, whether people want it to be or not. |
||||
|
||||
|
AsleepBySunset
Posts: 220 |
|
|||
|
I don't agree with the fatalism.
I don't actually believe AI programs like chatGPT and MJ are going to get free passes. Sam Altman recently remarked that he was willing to "go around firewalls to get more quality data", which he made some flimsy excuse for when called out, meanwhile microsoft is planning on incorporating AI heavily in windows 11 which will likely be used to extract "high quality data" from your computer/file system (via programs like onedrive) to train microsofts AI (bing) and microsofts business partner's AI (chatgpt). This is an invasion of peoples privacy which is being used to train AI's which have no ability to forgot; its literally an unsolved probelm in computer science. And at the end of the day, if the AI can't forgot and is trained on private data, licensced data, and frankly serves no utility to society aside from automating the arts and student plagiarism, then I think governments will force total algorithmic disgorgement of AIs like chatgpt. And either way, I'm probably going to make my own "AI free media" database so I can avoid garbage like this. |
||||
|
||||
|
moozooh
Posts: 150 |
|
|||
|
Windows 11 telemetry and other Microsoft shenanigans have literally nothing to do with training large language models like GPT. They're trained on public data: books, articles, random webpages. It wouldn't even make sense to specifically invade users' privacy for something that could be scrubbed off of any public web space where people talk to each other like this forum right here.
Other than that, I don't remember a single historical precedent when governments would purposefully and successfully hinder progress in automation as opposed to encouraging it. Sounds completely backwards. Besides, companies aren't going to stop developing AIs even if you ask them nicely. Worst-case they'll just move their HQ to another country with a more agreeable government. AIs becoming better than humans at a given task is the exact same kind of problem as a car running faster than a horse: not something you can prevent—but something you can take advantage of if you can use it instead of a horse (and aren't a horse yourself). Whether you're pro-AI or anti-AI, you need to keep a level head and not give in to wishful thinking. But anyway, good luck with your AI-free media database, whatever that is. Perhaps you'll be able to somehow take advantage of it while AI output is still distinguishable from a human's. |
||||
|
||||
NeverConvex
 Subscriber Posts: 2356 |
|
|||
It's not a solution for targeted forgetting, but to avoid privacy issues in the first place, LLMs could be trained using some form of differentially private variant of stochastic gradient descent (with modest small budget). There's at least one reasonably careful paper that does this in computer vision (convolutional neural nets), but I don't know of any that have tried it for LLMs. It entails a potentially significant performance decrease, so some significant pressure from privacy advocates would be needed to get the AI community to take it very seriously in large-scale production use.
I think they're normally trained on publicly accessible data, but I'm doubtful they're very careful about IP protections on that data -- i.e., the difference between publicly accessible and publicly owned may be quite important here. Last I checked, though, there's not really any clear case law on to what extent AI trained from massive input datasets owes anything to the creators of the initial training data. Anyone's guess as to how that could evolve, IMO; maybe class-action lawsuits alleging use of training inputs without permission will become a big thing, the more traditional industries are disrupted by ANNs (whether LLMs or CNNs or whatever else, depending on domain). |
||||
|
||||
|
moozooh
Posts: 150 |
|
|||
Yeah, this is really the key issue. On the one hand, the way regular humans are trained on other people's intellectual property is well-understood and doesn't make them liable for anything as long as the similarities between their product and someone else's aren't obvious and proven beyond reasonable doubt. On the other, the ethical considerations of using e.g. ArtStation samples for training in order to specifically imitate what's trending on ArtStation sounds like a breeding ground for plagiarism. I can see commercial marketplaces closing off their samples to avoid scrubbing, but let's be honest here: unless there is a need to specifically imitate the style of something made recently by specific artists, there is more than enough publicly accessible data—licensed permissively or otherwise—that it would realistically be enough to train a model of virtually any scale. As I see it, just thinking in terms of what's legally feasible to process, it's likely that legislation will move toward what's currently used to assert regular human plagiarism, i.e. if similarities are found and proven in the final product, and the rest will be left to the discretion of individuals. |
||||
|
||||
Meexa
 Posts: 186 |
|
|||
Nah I am ready. I'm tired of looking at my top 10 myanimelist and top 10 imdb list, with the thing hardly budging. I'm tired of going through movie after movie, and anime after anime, not finding anything good. Same with reading material. I'm also tired of working for people who are less kind and smarter than us. If AI can get better at creativity, which it will one day with AGI and quantum computing, this will be a better world to live in. Just right now, it isn't there, and I don't know if we will get it in our life time. But future generations definitely will. I believe the only thing our current generation is going to see and is seeing right now, is using tools like chatgpt to improve their creativity. We currently already use technology to do that already. A lot of people get too enthusiastic about our current tech. Just like how 3d has been reborn and died, multiple times. Everyone going like, "omg this is the next best thing and it will take over the world" for it to die out in less than a generation. Don't worry folks, it will get there, just not now. Lastly, those 2 ideas chatgpt gave for the next chapter, was not lit. I rather read Record of Ragnaroks next chapter. In fact, Record of Ragnarok needs to go bi-monthly and I want netflix to hurry up with part two of season 2. spoiler[Just so I can see hades vs qin shi huang.] |
||||
|
||||
 |
All times are GMT - 5 Hours |
|
|
|
Powered by phpBB © 2001, 2005 phpBB Group