ELI12 start: the LLM track in plain language
ELI12 = "explain it like I'm 12." Each ELI12 note retells one week with everyday pictures and explains every symbol before using it. These notes are a companion to the full notes, not a replacement. Exam wording lives in the exam-focused guides and the past-paper banks, and where the two differ, use the exam wording.
Tip
Suggested route. For each week: read the ELI12 note, then the full notes, then the exam-focused guide, then try the past-paper questions.
The four notes
| Week | Plain-language note | One-line big idea | Full notes |
|---|---|---|---|
| 1 | Week 1 in plain language | Attention lets the writer look back at every source word instead of one squeezed summary. | Week 1: From Sequence Models to Transformer Encoders |
| 2 | Week 2 in plain language | A decoder writing word \(t\) may use the source and its own earlier words, never future words. | Week 2: Transformer Decoding, Position, Normalization, and Training |
| 3 | Week 3 in plain language | A language model gives a probability to every possible next word; GPT-1 pre-trains, then fine-tunes. | Week 3: Language Modelling, GPT-1, and Transfer |
| 4 | Week 4 in plain language | A decoding rule turns next-word probabilities into text, and BERT learns by filling blanks. | Week 4: Decoding Strategies and Bidirectional Language Representation |
Suggested reading order
- Start with Week 1 in plain language: from one squeezed summary to attention and the Transformer encoder and work upward. Each week builds on the one before.
- For Week 1: Week 1: From Sequence Models to Transformer Encoders, then Week 1: Attention and Encoder Exam Guide, then Week 1: LLM Exam Questions and Worked Answers.
- Repeat the same sequence for Weeks 2, 3 and 4 using the matching files.
- Return to the LLM Knowledge Map whenever a concept link is unfamiliar.
Note
What these notes are not. Worked answers in the banks are derived study aids, not official answer keys. The ELI12 notes simplify the language, not the facts.