The core-concepts thread

The exam's dense definitional vocabulary — next-token prediction, tokenization schemes, embeddings, transformer building blocks, prompting techniques, evaluation metrics, benchmarks and trustworthy-AI principles — the named terms tested directly and reused as building blocks everywhere else.

T6 · 57 lessons across 8 modules

1. M0101-01Next-token prediction: what a language model is trained to do
2. M0101-02LLM parameters: what they are and where knowledge is stored
3. M0101-03Tensor shapes in transformers: batch, sequence, and hidden size
4. M0101-04Vectors, dot products, and cosine similarity
5. M0101-05Loss functions and cross-entropy explained
6. M0101-06Gradient descent and backpropagation
7. M0101-07Train, validation, and test splits
8. M0101-08How to build an evaluation set for an LLM project
9. M0202-01Why text must be converted to numbers
10. M0202-02Tokens, vocabulary, and subword tokenization
11. M0202-03Counting tokens: why tokens are not words
12. M0202-04BPE vs WordPiece vs SentencePiece: which model uses which
13. M0202-05Stemming vs lemmatization, and stop-word removal
14. M0202-06Bag-of-words, TF-IDF, and n-grams
15. M0303-01What text embeddings are: learned dense vectors
16. M0303-02Token embeddings vs sentence and document embeddings
17. M0303-03How to choose an embedding model
18. M0303-04How to test retrieval quality by hand
19. M0303-05Vector arithmetic and word analogies (word2vec)
20. M0404-01Self-attention and why context length costs quadratically
21. M0404-02Positional encoding: how transformers represent word order
22. M0404-03Encoder-only vs decoder-only vs encoder-decoder models (BERT, GPT, T5)
23. M0404-04Autoregressive generation: how an LLM produces text
24. M0404-05Temperature, top-k, top-p, and greedy decoding
25. M0404-06The context window: what it is and how to budget it
26. M0505-01Zero-shot vs few-shot prompting and in-context learning
27. M0505-02How to structure a prompt: instruction, context, and format
28. M0505-03Chain-of-thought prompting: when it helps and when it misleads
29. M0505-04Prompt templates, versioning, and testing
30. M0505-05Getting structured JSON output from an LLM
31. M0505-06Prompt, RAG, or fine-tune? A first decision rule
32. M0909-01Scaling an evaluation set to a hundred items
33. M0909-02Perplexity: what it measures and what it misses
34. M0909-03Human evaluation: rubrics and inter-annotator agreement
35. M0909-04BERTScore and embedding-based evaluation metrics
36. M0909-05How to choose an evaluation metric: loss functions, R², precision vs recall
37. M0909-06BLEU vs ROUGE vs exact match: which metric for which task
38. M0909-07RAG evaluation metrics: faithfulness, relevance, and context recall
39. M0909-08Cross-validation: k-fold, stratified, and when not to use it
40. M0909-09Sample size and statistical significance in LLM evaluation
41. M0909-10LLM-as-a-judge: how it works and where it fails
42. M0909-11Reproducibility: why temperature 0 is not deterministic
43. M0909-12Why LLMs hallucinate, and the types of hallucination
44. M0909-13Error analysis: turning failures into a fix list
45. M1010-01Public benchmarks (GLUE, MMLU) and data contamination
46. M1010-02Zero-shot and few-shot capability testing
47. M1010-03A/B testing an LLM feature in production
48. M1010-04Regression testing an LLM system in CI/CD
49. M1313-01NVIDIA's four pillars of trustworthy AI and how to implement each
50. M1313-02NVIDIA NeMo Guardrails and content moderation for LLM applications
51. M1313-03Prompt injection and indirect injection through RAG
52. M1313-04Bias in AI: sources, measurement, and mitigation
53. M1313-05Data privacy, consent, and why model weights cannot forget
54. M1313-06Transparency, explainability, and model cards for LLM systems
55. M1313-07AI energy use, carbon, and efficient inference
56. M1313-08A trustworthy AI checklist for your own LLM service
57. M1313-09How to read an AI research paper and track new LLM trends

Part of the throughlines running across the NCA-GENL prep course.