

Mr. Chatterbox
#22 v Konverzační AI a chatbotiTrip Venturella · 8× · naposledy 30. 6. 2026
Mr. Chatterbox is a language model developed by Trip Venturella with approximately 340 million parameters, trained from scratch using Andrej Karpathy's nanochat framework on a corpus of over 28,000 Victorian British texts (1837–1899) from the British Library dataset. The model is available as a 2.05 GB file on Hugging Face (Model repo "tventurella/mr_chatterbox_model", MIT license) and can be used via a Hugging Face Space demo in the browser or locally through CLI/plugin (e.g., llm-mrchatterbox). It is designed to exclusively reflect knowledge and linguistic style from the 19th century, with synthetic dialogue data from Claude Haiku and GPT-4o-mini additionally used for Fine-Tuning; the response quality is assessed by testers as still weak/b