George Keller Hart: “Large language models are potentially our civilization’s self blinding. How? By undermining our 5,000+ year heritage of written language – specifically because LLMs substitute memetics for attribution, breaking the first rule of shared, collective knowledge. Our vast corpus of recorded language has always rested on the provenance of each text, whether books, laws, correspondence, etc. – and across media formats as well. Attributed provenance is the foundation for evaluation and discernment, as opposed to LLMs that generate probabilistic ‘mimema’ – Greek for ‘imitations’.” Link