The most unsettling thing about AI is not how fast it is progressing: it's that its creators don't know how it acquires emergent capabilities

Top AI figures, including leaders at Anthropic and researchers like Geoffrey Hinton and Chris Olah, say they still cannot fully explain how large models develop emergent abilities.

The article weighs whether this uncertainty is just a marketing narrative, but points to interpretability research suggesting the gap in understanding is real.

That matters because if advanced systems gain new capabilities in ways researchers cannot explain, it complicates safety, control, and reliable deployment.