AI Unwrapped! How a Language Model Actually Works
6:41Coming soon
You ask a chatbot a question and it answers in confident, well-written sentences. Sometimes the answer is exactly right; sometimes it is made up, in the same confident sentences. To know when to trust it, you need to know what it is actually doing. This lesson is the real mechanism, with real numbers and no metaphors.
You'll learn
- Tokens: how text becomes numbers, with real token IDs from a real tokenizer
- Why models are limited and billed by tokens, not words (about four characters per token in English)
- Embeddings: each token becomes a long list of numbers
- Next-token prediction: a score for every token in the vocabulary, one picked, repeat
- Temperature, and why the same question gets differently worded answers
- Training versus inference: the weights are frozen when you use the model
- The context window: everything the model can see for one answer, and why long chats "forget"
- Hallucinations: why fluent and confident is not the same as true
- The real thing: a tokenizer page, search on and off, the same question twice
- Exam corner: naming a made-up answer correctly
Chapters
- 0:00Confident and wrong
- 0:40Tokens: text becomes numbers
- 1:41Vectors: each token becomes a list of numbers
- 2:10Prediction: score every token, pick one, repeat
- 2:56Training and inference
- 3:44The context window
- 4:24Why it makes things up
- 5:17The real thing: tokenizer and chat
- 5:45Exam corner
- 6:08Recap and next
Exam objectives: AI concepts and risks as they appear in CompTIA Security+ V8 and the draft Network+ N10-010. Objective numbers will be added once CompTIA publishes them.
Byte City is not affiliated with or endorsed by CompTIA.