BelGPT is a GPT-2 AI from OpenAI. GPT-2 has been trained on 40 GB of text (equivalent to about 400 times the Harry Potter saga), 8 million documents and 45 million web pages. But it only generates texts in English. To teach it French, the corpus was replaced by a compilation of web pages, books and newspaper articles in our language.
AIALAB: Text was created by specializing the corpus used in the desired styles.
The inspirations are diverse:
– The grimms brothers
– Émile Nelligan, Arthur de Bussière, Nérée Beauchemin, Alphonse Beauregard, Blanche Lamontagne-Beauregard, Louis Dantin et William Chapman
– Claude-Ignace Brugière, baron de Barante, Edme Boursault, Molière, Thomas Corneille, Pierre Corneille et Jean du Mas d’Aigueberre
– Mylène Fortin