Chat-GPT Commentary/AI modelled commentating
Something I would pay a premium for... is actual option for commentary. Use AI modelling to get it to commentate based on data from crofty/others to speak in a manner like them and license the commen...
Forum Discussion
@movesmoreair This is not a good idea. This would be problematic for multiple reasons, first of all, all generated content would require manual validation to avoid explicit content or content that can be considered offensive, it's not as simple as just generating new content and shipping it, because no AI model are flawless, and adding something like this would most definitely alter the age rating as they wouldn't be able to guarantee that everything that is generated could be considered "safe".
But most important of all, there's big legal implications of generating voices and the ownership of it, I do not think that there would be anyone that would license out or the give the rights to do this, as you first need to sample, and then train a model of the voices.
Who owns the voice recordings of the commentators, who owns the rights to the likeness of the commentators, who owns the model that is trained on the real voice? Who owns the generative content?
As you can see, it would quickly become a legal limbo and/or minefield. As far as I know, no legal framework currently has support for AI content like this, so to explore this option would potentially be very costly.
Not only would you be able to make the generated voices say things that the owner of the original voice actor doesn't agree with or stand by, you would also make the commentators and voice actors redundant.
So it is safe to say that this will never happen, and I hope it never will as you would get lifeless impersonations of "the real thing".
"AI" is not the solution for everything.
As a software developer (specialized in Artificial Intelligence) you are really wrong about a lot of the things you said. First of all, since you have to train a model, if the model doesn't have any have any offensive material/word, the model will not be able to produce offensive word. So yeah, you can control the words and what the model will generate.
Second, you talk about the voices: there are two possibile solutions. The first one is not using the voice of the commentators we know, but hire some voice actor. The second one is buying the ownership of that voice... yeah read it right: this is happening right now, where stars like snoopdog sells their image/voice to AI models.
So yeah, AI is the solution for a really dynamic and innovative commentary in Fifa. Hope they do this really soon.
probably you should also know that for LLMs you do not train a model from scratch, you fine tune it and you have no control on how the data used to train the base model ... sure they could just train their GPT model from scratch but it depends on their budget .
