AI City: Gen AI and Agents in 3D

In simple words: An AI model learns the way you learn a language: by reading a huge amount of it. Before training, companies collect trillions of words from books, websites, code and conversations, then clean out spam, repeats and private details.

In simple words: Each token is turned into a list of numbers called an embedding. Think of it as an address on a giant map of meaning: words with similar meanings live close together, so "king" is near "queen" and far from "pizza".

In simple words: Training means showing the model a sentence, hiding the next word, letting it guess, and nudging billions of numbers (parameters) so the next guess is better. Then fine-tuning and human feedback (RLHF) teach it to be helpful and safe.

In simple words: A model only knows what it saw in training. Retrieval-Augmented Generation (RAG) first searches your own documents, picks the most relevant pieces, and gives them to the model with the question, so answers come from your data with sources.

In simple words: An agent is an AI that works in a loop: it plans the next step, uses a tool, looks at the result, and decides again, until the job is done. Watch the little bots run between HQ and the workshops.

In simple words: The Model Context Protocol (MCP) is an open standard for connecting AI apps to tools and data. Build a connector once as an MCP server, and any MCP app can use it: ships from every island dock at the same port.

In simple words: Evals are exams for AI apps: a fixed set of test questions with automatic checks. Run them after every change to know if the app got better or worse, and trace every step to see where it went wrong.