Is AI really new? A history of ideas becoming useful
A brief journey through the ideas behind artificial intelligence, what changed along the way, and why a new capability is not the same as a reliable system.
Blog / Pedro Acosta
Understanding AI.
Explanations and reflections on artificial intelligence, from its foundations to the systems we build with it.
AI explained
Recommended reading order
What is a neural network?
From a handwritten digit to a prediction: how neurons and layers work, how a network learns, and how its architectures fit into artificial intelligence.
From words to vectors: tokens, embeddings and why meaning becomes geometry
Follow a sentence from tokens to identifiers and learned vectors, then see why context is still needed to interpret a word.
What is attention, and what is it for?
Follow an ambiguous word through queries, keys and values to see how attention combines information from context, and what its weights can tell us.
What is a Transformer?
Follow the fisherman from an English sentence to its translation and discover how a Transformer connects information to build an output, without needing formulas.
What does a model learn from? Training data
Follow a collection of handwritten digits into a text corpus to see how sources, selection, duplicates and missing examples shape learning and evaluation.
What is an LLM, and how is it trained?
Follow a fisherman's sentence through predictions, loss and parameter updates to understand what pretraining changes, why scale matters and how learning is checked.
AI FAQs
What is a decision model?
A broken mug connects typed AI answers, calibrated probabilities and the cost of acting. Why a likely answer does not choose the action for you.
Do more parameters make a better LLM?
What 27B really means, why total and active parameters differ, and how to read context, memory and reasoning limits without mistaking a specification for a quality score.