Take the sentence “The bank approved the loan because it<\/strong> had strong collateral.” Attention is how the model works out that “it” means the loan and not the bank. It gives a strong weight to “loan” and “collateral” when interpreting “it.”<\/p>\n Bias testing<\/strong> means checking whether an AI system treats different groups of people unfairly or produces systematically skewed results.<\/p>\n For example, suppose an AI system is used to screen job applicants. Bias testing might compare its recommendations across gender, age, ethnicity, disability status, or other relevant groups<\/strong> to see whether one group is rejected much more often without a valid job-related reason.<\/p>\n It can involve checking:<\/p>\n Bias testing asks: \u201cDoes this AI work equally well and fairly for different groups of people?\u201d<\/strong><\/p><\/blockquote>\n Deep learning<\/strong><\/span> is machine learning that uses multi-layer neural networks to automatically learn complex patterns from data. It is called \u201cdeep\u201d<\/strong> because the neural network has many processing layers between the input and output. Input, hidden layers, output<\/p>\n For example, with image recognition:<\/p>\n Deep learning is used in things such as ChatGPT and other LLMs, speech recognition, image recognition, translation, autonomous driving, and medical image analysis<\/strong>.<\/p>\n Embedding<\/span> <\/strong>– a model can convert text into an embedding<\/strong>, which is basically a long list of numbers representing its meaning.<\/p>\n Foundation model<\/strong><\/span> -is a large, general-purpose model trained once on a huge, broad dataset of text, code and sometimes images. It is then adapted to many different tasks instead of a new model being built for each one. Examples are GPT-4, Claude, Gemini and Llama.<\/p>\n The old approach was one model per job: one for fraud, one for document classification, one for chatbots, each trained from scratch. A foundation model reverses that. You build one general model and adapt it in one of three ways:<\/p>\n Grounded<\/strong><\/span> answer is an AI response supported by trusted source data or retrieved documents.<\/p>\n Inference<\/strong> is the stage where a trained AI model is actually used to make a prediction or generate an answer<\/strong>.<\/p>\n simple way to think about it:<\/p>\n Training = learning<\/strong> For example, with ChatGPT:<\/p>\n For image recognition:<\/p>\n In an LLM system, the flow is roughly:<\/p>\n Prompt \u2192 model inference \u2192 generated response<\/strong><\/p>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n<\/div>\n Neural Network – <\/strong><\/span>a machine-learning model inspired loosely by the way neurons in the brain process signals.\u00a0It is made of connected units called artificial neurons<\/strong>, usually arranged in layers:<\/p>\n Input layer \u2192 Hidden layer(s) \u2192 Output layer<\/strong><\/p>\n For example, for recognizing a cat:<\/p>\n Each connection has a weight<\/strong>. During training, the network adjusts those weights to reduce errors and improve its predictions.<\/p>\n Goal – recognize Circle<\/b><\/p>\n Wrong prediction, needs to be trained<\/b><\/p>\n Neurons that learn patterns by adjusting weights during training.<\/strong><\/p>\n And the relationship is:<\/p>\n AI \u2192 Machine Learning \u2192 Neural Networks \u2192 Deep Learning<\/strong><\/p>\n Deep learning<\/strong> simply means using neural networks with many layers<\/strong>.<\/p>\n Prompt injection<\/strong><\/span>\u00a0– is when someone hides instructions inside text an AI reads, and the AI follows those hidden instructions instead of (or as well as) the ones its owner gave it.<\/p>\n RAG (Retrieval-Augmented Generation) –<\/strong><\/span> an AI technique that retrieves relevant external information and gives it to a language model so it can produce more accurate, grounded answers instead of relying only on what the model learned during training.<\/p>\n\n
\n
\n
\nInference = using what was learned<\/strong><\/p>\n\n
\n
\n
![\"\"](\"https:\/\/davidpapkin.net\/wp-content\/uploads\/2026\/10\/image.png\")
<\/p>\n![\"\"](\"\/wp-content\/uploads\/2026\/10\/image-1.png\")
<\/p>\n