Limitations of current copyright frameworks for large language models trained on scientific literature | Synapse