Large language model papers explained

Language models learn useful representations by predicting missing or upcoming text. That simple objective leads to different capabilities depending on the training data, model design, and how the model is prompted. These explainers connect the training objective to the behavior you see when you use a model.

Start with two contrasting approaches to pretraining, then explore how retrieval gives a model information beyond its weights. The collection also covers prompting, compression, and the systems that make inference practical.