Oh hell yeah

You're viewing a single thread.

73 comments

Too bad the supreme court put the dagger in artists backs for AI already. It would have been great if a company would license an artists source material to make a set of variations for a limited venue. Like a company licensing a voice actor's samples, then they get to use AI to make those characters say whatever they want, in that one movie, or that one game, based on the license.
As it is now, we're going to end up with Spruce Lee fighting Hackie Chan movies, and none of the actors or their estates will get to say shit about it.
- You don't really know how AI works, do you? A single voice actor couldn't produce enough lines to fully train an AI model even if they spent every second of their life in the recording booth.
  So tell me then, which of the billions of input recordings do you pay licensing fees for and how much? I mean sure, we could make a law that forces AI companies to pay for every single piece of training data. Which would probably kill AI training for the entire region where this law applies, severely crippling our already weakened economy. But I guess, at least we're keeping the moral high ground while doing so.
  But seriously, the EU is cooking up a pretty amazing Ai law right now. Thought out by people far brighter than you and me and it seems to be pretty amazing at balancing economic interests with ethical obligation. My hopes are high for that one!
  
  AI deepfakes are all over the place right now. Or did you think the video of Trump being a rock star was real?
  
  Ok? And what does this have to do with anything I wrote?
  Edit: Oh, I get it. You misread the first paragraph and completely ignored the second which made you completely miss the point of my comment.
  I'm not saying that you can't specialize AI to sound like a specific actor. I'm saying that you can't train a new AI model using only recordings of a single actor.
  
  You can record a few seconds of dialogue and get convincing results. I’m sure at a professional level, you could do it with a few minutes at most.
  (Side note, I don’t endorse or recommend the site I linked to. Only using it to demonstrate the point. I don’t recommend you surrender the sound of your voice to these guys.)
  
  You can license the output of a likeness powered by AI.
  
  what confuses me a lot about america is how a lot of people will defend the AI's "right" to steal training data to learn (education)
  when there's millions of students out there going into life long debt to go to collage, and none of the same americans are fighting for their right to learn for free
  
  I am only critical that those models are in the hand of corporations who try to profit from it. Copyrights are mainly a tool to be wielded by the powerful: see Sony trying to disconnect ISP accounts en masse or media giants suing people into oblivion, Nintendo fucking over their fanbase again and again and so on.
  The datasets should belong to an UN organisation like UNESCO, corporations/NGOs/people should be able to licence them to build their models (ev. with "community models" provided free for personal use), and the licence fees should be used to subsidize culture. This plus an UBI would make sure that artists don't have to starve, corporations can use them to try to make a profit, and everyone else can use them to create for their own or their communities use. Artists that don't want to go into the datasets have that right too, but also won't have access to that financial pool (this shouldn't be the only pool).
  Fuck copyrights.
  
  If we just got rid of copyright, we would just see Disney flipping other people's work without credit, just like how genAI grifters try to use it currently.
  
  Under capitalism, copyrights are a necessary tool for people to get paid for their labour in some cases. That's not the most common usecase for copyrights, but a needed one until we get a better system. Fuck copyrights being used for corporate greed though. Somehow, more people are against the former than the latter, which is super wrong. But it's not a reason to be against the former in addition to being against the latter.
  
  The fastest fix is a very short copyright window, say 5-10 years. Given the fact that every movie release makes back it's entire production budget in 1 week or is immediately deemed a flop, 5-10 years is plenty of time to mine an IP for boatloads of cash, plus once anyone can make a star wars movie or book or whatever, the real value shifts instead to the brand that produces it. New brands can make a name for themselves respinnong "old" IPs and making far better content than the original studio would've been willing to risk, and big brands can focus more on how important it is to view the only official versions of a given IP based on their branding.
  
  That'd also harm big corporations less than small, independent artists who don't get paid as much per use of their artwork.
  
  1st, I'm not Murrican, I'm German. 2nd the right to free education is one of the most important human right and the entire world is paying the price right now for neglecting this right over the past decades 3rd AI is here to stay. It's far to impactfull as a technology. Whether we like it or not. So we either create an Environment where it can thrive within certain rules or we watch as others use it to completely overtake us.
  
  ok? and i'm polish, i was talking about americans.
  AI is not just ethically dubious, it'd also outright harmful for the already strained environment we live in. If AI stays - it won't be here for long, mostly because there won't be anoyone to ask it to generate giant hentai tits anymore
  
  AI is here to stay. It's far to impactfull as a technology
  Gonna need some proof for that. So far it doesn't actually do anything useful.
  
  Open your eyes and step off that hate bandwagon.
  Machine Learning has revolutionized protein folding and plenty of other sciences. LLMs have increased programmer productivity (even if it isn't perfect yet). Image/video/song generating was something we thought to be impossible a couple of years ago.
  If the only news you get about AI comes from the "Fuck AI" community, you won't ever get accurate info.
  Yes companies put AI in a bunch of shitty things that don't need it. But to claim AI doesn't do anything useful is just plain wrong.
  
  Machine Learning has revolutionized protein folding and plenty of other sciences.
  I actually work in the field of protein crystallography. Contrary to newspaper reporting by people who don't understand the field and just repeat what the people who developed the tool say about it, it has made just a small improvement to analysing experimental data which we could have easily made using traditional algorithmic approaches with a similar amount of resources spent. And this is one of its biggest legitimate impacts - it absolutely hasn't "revolutionised plenty of other sciences", or you'd be able to list more things than just alphafold.
  It doesn't improve programmer productivity, it increases the lines of code created, which is a really bad metric for productivity. There is good evidence that its use is already leading to increased code churn, that means someone is having to go back and revisit the additional new errors introduced by AI tools, which is obviously less productive.
  
  So what you are saying is that AI is actually useful since it has improved analyzing experimental data.
  Thanks for proving my point.
  
  As a software engineer, I can tell you that it absolutely has increase productivity. Especially for small tasks without too much complexity. AI is really good for prototyping. The problems you hear about are mostly people who have no idea how to write propper code trying to mask their incompetence by writing AI code.
  I usually outright reject code that is obviously AI. But I employ plenty of AI in my own coding. The trick is to always double check and rewrite segments that aren't good enough.
  The huge amount of garbage AI PRs ate an enormous problem. Especially for small open source projects. But the benefits are also pretty obvious.
  
  A single voice actor couldn't produce enough lines to fully train an AI model...
  The model is trained on a massive corpus of existing data and then fine tuned to match the target voice actor. Using less than ~30s of reference audio you can get a pretty decent fine tuning the main issue is that it currently isn't on par with the quality and consistency of an in studio voice actor, especially over long time domains.
  
  Hence my usage of the words "fully train". The other commentor wants to license every piece of audio used in training the model which obviously includes the base model...
  
  You can feed an infinite amount of data into existing models and it won't improve the issues. The problem is with the models themselves.
  And the audio used to train the base model are licensed. Usually under an MIT, creative commons, etc. license.
  
  In Spain we trained an AI using recorded congress sessions. Within the national, regional and city halls they had a lot of material.

73 comments