Raw LLM Responses

This video is filled with misinformation about how AI learns and generates content as well as false claims about copyright protections. To start with AI learns by having an image deconstructed down to a random assortment of static noise. The AI then learns how to construct an image from that noise based on criteria from its learned understanding of things that share common features. So an AI told to generate an image based on an artists style will have criteria developed on the similarities between different images that artist has done and it will generate an image that shares those qualities. It does NOT copy elements from any artist's actual work. It does not have a database of millions of images. A dataset is the criteria for what the common features are that are associated with the terms used in prompts. So in a very basic example if you put in a prompt for an apple the dataset does not have a library of images of apples to copy and paste from, it has a series of things that apples have in common like an apple may be round, an apple may have a stem, an apple may be red, an apple may be green, an apple may be yellow, etc... This extensive list of criteria is associated with probabilities such that the more things that match the higher the chances are that the thing in question is an apple. Humans do in fact recognize things in the same way. So a claim that AI is copying artists is flat out wrong, it creates images by recognizing elements from the prompt it was giving within a random distribution of static noise and tries to reconstruct an image from those recognized elements. Even when an AI is trying to reconstruct an image used in the training process from the deconstructed static it doesn't reconstruct it perfectly because it's not using the original image in the reconstruction. It's using it's understanding of the common elements of a given prompt. As for copyright protections, copyright protects the specific works created by the artist. So if an AI is used to very closely resemble an image that an artist made then that specific instance is a violation of copyright protection and is actionable. However, an artist's style is not protected under copyright. So creating a completely unique piece of art that looks similar to how a specific artist may have rendered that idea is not a violation of copyright protection. To the point of how you can and can't use something that is freely viewable online copyright law does offer certain protections in this regard being that you can't necessarily take an image an artist posted to their publicly viewable gallery (or other publicly viewable repository) and use it in something like your advertising campaign or to put it on a t-shirt to sell. This would be taking a direct copy of their work and profiting from it. You also can't take their work and modify it a bit and then use it because that falls under it being a derivative. However, derivatives are allowed under copyright if enough of the original has been altered to be considered something new in itself. Regardless of such allowances though AI is doing neither of these things, copying directly or creating a derivative. It is creating something new from a random distribution of noise. Its use of publicly viewable art to create its dataset of criteria is no different than a human going onto an artist's repository and studying their art to learn how to create art in that style. Such use of publicly viewable art is not a violation of copyright protections for a human to do so there's no reason for it to be a violation for an AI to do it either. Thus to claim that there was theft involved or that artists should receive compensation because an AI learned from their publicly viewable repository is to claim that all artists stole from other artists or that they should compensate the artists that they learned from. To the matter of the artistic process, when a person generates an image through the use of AI they rarely get something that matches what they wanted with a single prompt. It takes lots of revisions to work out what will get results close to what you actually want. Often times you even then have to alter what was generated either by using something like photoshop or using the generated image as a base while adding in new prompts to generate alterations to the initially generated image. All of this is the human element being added into the machine element of the AI. So to make a blanket claim that AI art is not art is a bold claim that devalues the work that humans do put into the creation of the art they make with the help of AI. Can AI be used to generate soulless images that took no real effort, yes, but artists can and do put out the same types of things. It's in how its used.