Can sophisticated AI models predict the future? Many say yes, but computer scientists say that public language models (LLMs) are just algorithms that can’t think. They can only regurgitate existing text based on training data.
Using this hypothesis, LLM Talkie was trained with 13 billion parameters. Talkie is unique among AI models because it was only trained on data from before 1931. Why 1931? Because all works before 1931 have entered the cloud domain, there are no ethical issues in training AI models using copyrighted works.
Talkie was developed with the aim of understanding how models predict the future without modern data. It will also be used to see how pre-internet knowledge and culture shape the model’s behavior. In addition, studies are being conducted to see how far the model can think outside its training data.
Apart from experimenting with LLM's ability to predict the future, Talkie can also be used for entertainment purposes to provide the experience of talking to individuals from the past in the 1930s, complete with old language and vocabulary. It can generate dialogue, essays, and classic Victorian and Edwardian writing styles. Accurate history of the world in the context of the 1920s can also be done.
Talkie can be accessed via chat sites, Github, and Huggie Face today.
