LLM Tokenization Explained: How Language Models Split Your Text Into Tokens
How tokenizers turn text into IDs, word vs character vs subword tokenization, how byte-pair encoding builds a vocabulary, why tokens shape prompt limits
Computer Vision
AI in Healthcare
Train Validation Test Split: How to Divide Your Data Without Fooling Yourself
What the training, validation and test sets each do, how to choose random, stratified, grouped or time-based splits, when to use cross-validation, and
AI Ethics
Vector Embeddings Explained: How AI Turns Words and Images Into Comparable Numbers
What vector embeddings are, why they replaced one-hot encoding, static vs contextual embeddings, a cosine similarity example you can check by hand, and
Blog
Model Cards: How to Document a Machine Learning Model So Others Use It Safely
What model cards are, the research behind them, a seven-section outline, how Hugging Face README model cards work, a short worked example, and
AI Ethics
Differential Privacy in Machine Learning
Dwork and Roth define differential privacy as a definition. Machine learning is a later application in the monograph, not a badge and not
Artificial Intelligence and Technology News
Inverse Kinematics Explained: How Robot Arms Work Out Their Joint Angles
Forward vs inverse kinematics in plain language, a two-joint arm solved by hand, analytical vs numerical solvers, and the joint limits, singularities and
Blog
What Is Cloud Computing and Edge AI
NIST defines cloud computing in SP 800-145 and, later, edge computing as the end-device layer. Neither document defines an edge AI product.
Robotics
Image Data Augmentation: Which Techniques Help Computer Vision Models, and Which Hurt
How image data augmentation works, the geometric and colour transforms worth using, automatic policies, how to apply them in TensorFlow and PyTorch, and
Machine Learning
AI Fairness Metrics Explained: Demographic Parity, Equalized Odds and How to Run a Fairness Check
What demographic parity, equalized odds and equal opportunity measure, how to choose between them, a four-step fairness assessment with Fairlearn, and the judgement
Machine Learning
How to Train an AI Assistant in JavaScript Without Pretending the Browser Is a Cluster
The search phrase covers three jobs: a prompt, a pretrained model in the browser, and a real training run. This note keeps them