Talkie: A 13B Model Trained Entirely on Pre-1931 Data
A new project named Talkie introduces a 13B language model trained solely on historical texts from before 1931, helping researchers study AI's generalization capabilities when faced with 'vintage' data.
Sources x.com