Training data for gen AI new features

Hi. I would like to start this respectfully because I am more than aware of how hot the AI related debate is still going on, the lawsuits against various AI companies, specifically those selling services with genAI models.

Some artists like myself are very concerned about how these models are trained and where the data comes from. Recently I have seen that Davinci is releasing some AI features, researching with some colleagues I know that not all AI is necessarily a problem, there are several categories of them like Discriminative AI, Neural Networks, Machine Learning, which are mostly some of the features that are now available in DR20 and has been with us without the “AI buzzword” in it.

However, two of them are gen AI features like AI voice converter and AI set extender, so I would like to know how these two gen AI features have been trained and where the data came from.

I ask this because some of us were relieved that Davinci had not implemented genAI like Adobe and so some of us decided to change our software to something more ethical and also because it is top notch software in the industry.

Also, I researched weeks ago that CEO Grant Petty said he will not add genAI to Resolve and their products because he believes our work belongs to us, also BMD cloud service states on their site that they will not use our data and work to train their AI models through their cloud services.

So, I know that genAI needs a large amount of data to train their models and that is why I am concerned now.

I repeat and clarify that I am not against all kinds of AI, I just have problems with genAI models, no matter if they are third party programs, or if a company decides to implement it on their own stuff.

Procreate and some other companies decided to side with the artists, so I would like some clarification on this issue and to what extent these models are reliable. It would be good a disclosure of transparency in this regard and also taking into account of how many artists worldwide from different areas are concerned about it.

I would like to be able to appreciate how BMD differentiates and distances itself from these other companies and their bad practices. I think Davnici is a great program and as a filmmaker and animator I have loved it.

Thank you for your time.

I'd like transparency on this topic, too.

Resolve is such a powerhouse, I would think their marketing team could be confident enough in their product that they don't feel the need to hop on the AI bandwagon.

As annoying as it is (jumping on the buzzword bandwagon, adding toy-like "features" that no one asked for), it's not a deal-breaker. Finding out it was trained on stolen data like so many other "AI" tools, very well might be a deal-breaker.

I'd like transparency on this topic, too.

Resolve is such a powerhouse, I would think their marketing team could be confident enough in their product that they don't feel the need to hop on the AI bandwagon.

As annoying as it is (jumping on the buzzword bandwagon, adding toy-like "features" that no one asked for), it's not a deal-breaker. Finding out it was trained on stolen data like so many other "AI" tools, very well might be a deal-breaker.

There's anything that we know about this topic lately? Do you have a recent info related to it?

I haven't heard anything.

Set Extender isn't available to even test yet. I expect there will be more information once it's available.

The AI Voice Convert feature does come with a few specific voices but otherwise the idea is that you will perform the training.

They probably bought the models from Nvidia. I get what you’re saying, but there’s something about AI I actually appreciate. No one really raises a fuss when machines replace factory workers or when farming became mechanised but now that AI is coming for jobs in the intellectual and creative space, suddenly it’s a huge issue. AI is democratising things and let’s be honest, the truly talented folks have nothing to worry about, it’s mostly the mediocre ones who might be out of a job. Funny how no one complains about how a robot was trained when it replaces someone on an assembly line.

As for training AI, I’m personally fine with it using stuff I’ve written or videos I’ve put out. If it’s public, it’s already out there for anyone to learn from. I get that IP laws exist but sometimes they feel more like a way to gatekeep access than to protect real creativity.