AI & Language — How Artificial Intelligence Learned to Speak

A FANNI BC × NODA educational event exploring how artificial intelligence evolved from statistical models and machine learning into modern large language models. The session covered key concepts including supervised, unsupervised and self-supervised learning; GPU and parallel processing; transformers, attention and self-attention; tokenization, embeddings and tensors; forward pass, loss functions, gradient descent and backpropagation. The discussion also explored the probabilistic nature of large language models, data bias, the quality of training sources, energy consumption, and the importance of human judgment in how AI is used.