Monte Carlo, temporal-difference learning, SARSA, Q-learning, actor-critic. The list grows quickly, and learning each name separately can hide what they share.
They all learn from sampled experience.
The new visual article follows one sample—state, action, reward, next state—through the learning loop. You will see how familiar algorithm families emerge from a few choices: when to update, whose behavior produced the sample, what to improve, and how to represent it.
Once that pattern clicks, an unfamiliar reinforcement learning algorithm stops looking like an isolated recipe. You have a small set of questions to ask.