Mr. Chatterbox is a (weak) Victorian-era ethically trained model you can run on your own computer
SMRTR summary
Trip Venturella released Mr. Chatterbox, a 340-million parameter language model trained exclusively on 28,035 Victorian-era British texts from 1837-1899 published by the British Library. The 2.05GB model represents an experiment in creating AI using only out-of-copyright material, avoiding the licensing issues plaguing modern language models trained on scraped internet data. However, the model performs poorly in conversations, feeling more like a basic text generator than a useful chatbot, likely due to insufficient training data compared to optimal ratios.
SMRTR provides this summary for quick context. The original article belongs to Daily.dev.
Read the original article