Does Predicting the Next Word Mean AI Understands? - Prediction, Error, and the Nature of Intelligence | Sleepy Facts About the Universe
From Sleepy Facts About the Universe · · 32 min
This plays from the publisher's own server. We do not host a copy.
Description
Got something to say to this podcast? Send a voicemail, and you might hear yourself on the next episode! (https://rss.com/podcasts/sleepy-facts-about-the-universe/voicemail)
Some questions sound simple until you try to answer them. This episode began with a deceptively straightforward inquiry: is intelligence simply prediction grown deep enough to question itself? To explore this, we look at how organisms from bacteria to fish use anticipation to survive, and how this capacity scales up to the complex counterfactual reasoning of a crow. The conversation then examines predictive processing, a framework proposing that the brain does not passively absorb sensory data, but instead generates a controlled hallucination, using prediction error to refine its internal models of the world.
We also turn to artificial intelligence, asking why large language models can fluently predict the next word without genuinely understanding the world. By distinguishing between shallow statistical extrapolation and deep model-based revision, the episode traces the boundaries of what makes a system truly intelligent.
I wanted to understand why a machine can generate a perfect sentence yet fail at basic physical intuition, while a tiny fish computes a life-saving future in milliseconds. Exploring this distinction reveals that the thread of prediction alone is not enough; the weave of flexible, revisable engagement with the world is what truly matters.