In a recent video from the Trust Issues Show, host Raji explains a fundamental truth about artificial intelligence: AI doesn't know facts—it calculates odds. This distinction is key to understanding why AI systems sometimes "hallucinate" or generate confident-sounding but incorrect information.
"AI is not a truth machine, it's a probability machine," Raji states. "Once you understand that AI doesn't know the truth—it knows the odds—hallucinations start to make complete sense."
Unlike humans who can reason about truth, large language models (LLMs) are trained to predict the most likely next word or token based on patterns in data. This probabilistic nature means that even when an AI sounds certain, it's simply generating the most statistically probable response—not verifying facts against a database of truth.
Raji breaks down why trusting a system designed to be merely "probable" rather than "right" leads to errors, and what that means for users who rely on AI tools for accurate information. The full episode is available on the Trust Issues Show channel.