factual

Attention entropy tracks human reading-time difficulty

The entropy of a transformer's attention distribution rises where linguistic theory predicts competing parses, such as grammatically incorrect 'were' agreement, and this matches longer human reading times, giving moment-by-moment model-human correspondence.

factualpending

Speaker

Chandra Sripada

Evidence Quote

“In LLMs, the attention... would be sharply positioned over 'problem.' ... there's confusion.”

Source

363 | Chandra Sripada on How LLMs and Humans are Cognitive Cousins— Sean Carroll's Mindscape
Created: 11/17/2026, 6:53:48 AM

Your notes

Loading notes...