‘Impossible’ to create AI tools like ChatGPT without copyrighted material, OpenAI says::Pressure grows on artificial intelligence firms over the content used to train their products

  • BURN@lemmy.world
    link
    fedilink
    English
    arrow-up
    3
    ·
    11 months ago

    I’m well aware of how machine learning works. I did 90% of the work for a degree in exactly it. I’ve written semi-basic neural networks from scratch, and am familiar with terminology around training and how the process works.

    Humans learn, process, and most importantly, transform data in a different manner than machines. The sum totality of the human existence each individual goes through means there is a transformation based on that existence that can’t be replicated by machines.

    A human can replicate other styles, as you show with your example, but that doesn’t mean that is the total extent of new creation. It’s been proven in many cases that civilizations create art in isolation, not needing to draw from any previous art to create new ideas. That’s the human element that can’t be replicated in anything less than true General AI with real intelligence.

    Machine Learning models such as the LLMs/GenerativeAI of today are statistically based on what it has seen before. While it doesn’t store the data, it does often replicate it in its outputs. That shows that the models that exist now are not creating new ideas, rather mixing up what they already have.