Raw LLM Responses

The reality is that AI companies don’t need access to copyrighted works at all to train models. In the latest paradigm, a language model trained only on public domain content generates an enormous volume of content used to train even better models. Also, the latest frontier models, like DeepSeek R2 don’t use transformer architectures at all, but far more sophisticated cognition architectures that can get smarter entirely from synthetic data. AlphaGo was trained on every Go game ever played and became really good at Go- however, the real breakthrough happened when they didn’t feed the model any information about real games but instead just allowed the model to play itself billions of times resulting in AlphaZero, which could consistently beat AlphaGo, and all human players, and as a bonus it invented entirely new strategies no human had ever thought of. The same is true of self-driving cars and humanoid robots. Initially they were trained on real world data, and they still use real world data for finding edge cases, but nearly all the training data is synthetic. So I wouldn’t worry too much about copyrighted works. The entire industry is outgrowing the need for access to any of it and is moving to synthetic training data. Also, the moment a paid commercial AI service like ChatGPT is released, a free open-source model like DeepSeek comes along that beats it and costs nothing. In the long-term, AI models for profit will never be commercially viable because a free model will always beat them. It’s all about what you build on top of those models that makes the real money. It won’t be long and this whole copyright issue will be a moot footnote in AI history. Soon, AI will be inventing most of the new breakthroughs, writing most of the popular books, generating most of the popular music, movies, and games. You will be able to turn on an AI music stream and listen to music in any genre generated just for you in the moment, such that you are the only person in the world who has ever heard any of those songs, most of which will only be heard once by a single individual. Every individual person will be able to generate an endless stream of unique media of any kind. At that point, copyright will become largely meaningless. AI will eventually make money and capitalism just as meaningless. You’ll see. 10 years from now, the world will be very different and it won’t be the oppressive techno dystopia you think it will.