How a language model works: from your question to its answer
A language model does not look things up in a database of facts: it predicts text from what it learned and from what it has in front of it. That is where its fluency comes from, along with the fact that it does not answer the same way twice and sometimes makes things up. This course opens the box in four blocks and without formulas, so that you know why it answers the way it does and where it is most likely to fail.
Private Course
Please sign in to contact responsible
| Responsible | Carlos Pascual |
|---|---|
| Last Update | 2026-10-06 |
| Completion Time | 4 hours 40 minutes |
| Members | 1 |
Share This Course
Share Link
Share on Social Media
Share by Email
Please login to share this How a language model works: from your question to its answer by email.
Basic
Data, Analytics and AI
-
-
Preview
-
-
Block 1: From text to numbers: tokens and meaning3Lessons · 1 hr 5 min
-
Chopping up the text: what a token is and why it matters
-
From token to vector: a map of meanings
-
Block 1 exam10 xp
-
-
Block 2: Attention: how the model uses what it has in front of it3Lessons · 1 hr 5 min
-
Attending: every word looks at the others
-
The context window: what it has in front of it and what it does not
-
Block 2 exam10 xp
-
-
Block 3: Training and adapting: what it learned, from where and up to when3Lessons · 1 hr 10 min
-
Learning to predict the next word
-
From base model to assistant
-
Block 3 exam10 xp
-
-
Block 4: Generating: why it answers this way and why it sometimes makes things up3Lessons · 1 hr 15 min
-
Word by word: sampling, temperature and different answers
-
Making things up with aplomb: where a false fact comes from
-
Block 4 exam10 xp
-