In this lesson
- It starts with a question. What is a language model actually trained to do?
- The idea you will find. Plausible and true are different targets. — A model is rewarded for text that reads like the real thing. Most of the time the most plausible continuation is also correct, which is why it is useful — and why the exceptions are so convincing. Treat its output as a draft to verify, not a source to quote.
- How you learn it. By doing: sorting.
- Length. 5 short steps. Progress saves on this phone.
Part of Inside AI Models
What a model actually learns from data, where it is strong, and where it fails.