What If We’ve Been Reading AI Wrong for Years?

What if some of AI’s most famous “uninterpretable” neurons were misunderstood? Nick Cammarata helped produce early research that made neural networks look like tangled combinations of unrelated concepts. Now, after revisiting those examples with better tools, he thinks we can understand much more of their internal machinery than he once believed. Leo Gao agrees that researchers sometimes gave up too quickly. But he challenges the conclusion: how do we know a convincing explanation actually describes what a neuron does? In this episode of Attention Span, we follow their disagreement inside the