A beginner-friendly introduction to how large language models work, covering core concepts including tokenization (how text is split into tokens), embeddings (how meaning is encoded as vectors), context windows, temperature and sampling strategies (top-k, top-p, beam search), prompt engineering, and zero-shot vs. few-shot learning. Also explains the difference between generative and discriminative models. First in a planned series building toward transformer architecture.
Table of contents
What is a large language model, really?How AI reads textEmbeddings are giving meaning a shapeHow much an AI can hold in its head based on context windowTemperature: The creativity dialControlling the word lottery though samplingHandling unknown wordsTalking to AI the right way through prompt engineeringPerforming without practiceA handful of examples goes a long way when learning via few shotsTwo philosophies of AIHow AI explore multiple paths at onceWhat you now knowSort: