Tech & Future
Understand the machines rewriting everything
Novice
Subdomain
The shapes behind models: vectors, gradients, probability, and loss. You don't need the proofs, just the right mental pictures.
Overfitting and generalisation
Memorising the practice exam isn't the same as understanding the subject.
Open
Transformer architecture
A neural network architecture built around self-attention layers that lets every token attend to every other token, replacing recurrence with parallel computation.
Open
Reinforcement learning from human feedback
A training procedure that fine tunes a language model with reinforcement learning against a reward model learned from human preference data.
Open
Parameter
A tunable weight inside a model adjusted during training.
RAG
Retrieval-Augmented Generation, giving a model external documents to ground its answers.
Overfitting
Memorising training data so well the model fails on new inputs.
Inference
Using a trained model to produce outputs.
Fine-tuning
Continuing to train a model on a smaller, more specific dataset to specialise it.
Embedding
A vector of numbers representing the meaning/features of a chunk of input.
RLHF
Reinforcement Learning from Human Feedback, alignment via human ratings.
Hallucination
A fluent, confident output that's factually wrong.
Token
The smallest input/output unit an LLM works with, typically 3 to 4 characters in English.
Context window
The maximum tokens a model can consider in a single forward pass.
Vector database
A store optimised for nearest-neighbour search over embeddings.
MoE
Mixture of Experts. Routes tokens to specialised sub-networks so active parameters per token stay small.