🧠 LLMs aren't technically Bayesian, but they're conditional probability powerhouses! They train with maximum likelihood yet output evolving token probabilities that update with each new piece of info. This visual captures how a neutral stock trend shifts up or down as new words enter the prompt.