AI & ML
From embeddings to agents — see how modern AI actually works.
AI & ML progress0 / 32
- What Is Machine LearningLearn the rule from examples instead of writing it.4m
- Training vs InferenceLearn once, slowly. Answer many times, fast.4m
- EmbeddingsMeaning turned into a fixed list of numbers.5m
- Cosine SimilarityCompare direction, ignore magnitude.5m
- TokenizationModels read chunks, not letters or words.5m
- Attention, IntuitivelyEvery position decides which others to listen to.5m
- Transformers: Big PictureA stack of identical blocks refining one sequence.5m
- What Is an LLMA next-token predictor trained on a lot of text.5m
- Temperature and SamplingOne distribution, many possible answers.5m
- Context WindowsA hard limit on what the model can see at once.4m
- Retrieval-Augmented GenerationFetch the facts first, then let the model write.5m
- Vector DatabasesNearest-neighbour search that stays fast at scale.5m
- Fine-Tuning vs PromptingChange the instructions, or change the weights.5m
- Agents and ToolsA model in a loop that can act and try again.5m
- Evaluating LLMsIf you cannot score it, you cannot improve it.5m
- Chunking and RerankingRetrieval quality is decided before the model reads a word.5m
- Adapters and LoRATrain a thin correction instead of the whole weight matrix.5m
- QuantizationStore the weights in fewer bits and buy back memory bandwidth.5m
- Approximate NeighboursWalk a graph of vectors instead of comparing all of them.5m
- LLM as a JudgeA model can grade answers — and quietly grade the wrong thing.5m
- The Tool-Use LoopThe model proposes a call; your code decides whether it runs.5m
- GuardrailsChecks around the model, because the model is not the boundary.5m
- BPE vs WordPieceTwo ways to decide which pair of pieces becomes one piece.5m
- The KV CacheKeep the past keys and values so each new token is cheap.5m
- Speculative DecodingA small model guesses ahead; the big one checks in one pass.5m
- Model Routing and FallbacksSend each request to the cheapest model that answers it well.5m
- Finding and Masking PIIFind personal data, swap it for placeholders, and mask the logs too.5m
- From Pilot to ProductionA pilot proves the idea. Production needs the system around it.5m
- AI Teams and the Center of ExcellenceOne central AI team, or builders inside every product team?5m
- Change Management for AIThe same tool can reach 15% or 70% of staff. The people plan decides.5m
- Data Readiness for AICheck what a use case needs from data against what actually exists.5m
- Measuring AI ValueBaseline first, then track value against cost until you scale, pivot or stop.5m
Quiz yourself: 3 questions from this module
1 / 3