general-purpose language model

Before LLMs, you usually built a separate model for each task: one for translation, another for spam detection, another for sentiment. A general-purpose language model collapses these into one. The same model translates, summarizes, writes code, answers questions, and brainstorms, just by being asked differently in plain language. You change the prompt, not the model.

This works because all those tasks can be expressed as text in, text out, and a model trained broadly enough on language has seen the patterns for each. The ability to switch tasks from instructions alone, with no retraining, is called in-context learning, and it is what makes one model a general tool. The flip side is that a generalist is rarely the absolute best at any single narrow task.