Meet Talkie-1930: A 13B Open-Weight LLM Trained on Pre-1931 English Text for Historical Reasoning and Generalization Research

Researchers released Talkie-1930, a 13B open-weight model trained only on pre-1931 English text.

The project also includes an instruction-tuned version and a live demo for experimentation.

It is designed to study contamination-free benchmarks, unseen-task learning, and temporal generalization.

The model can help probe how LLM behavior, persona, and identity shift when training data is fixed to a historical era.