It's difficult to say what thought is, but LLMs also solve more problems more correctly if you allow them to feed back through themselves multiple times (ask them to show their working) apparently for the obvious reason that it allows more computation to happen. I hypothesize that different people build different recurrence circuits, by chance. Some go through the audio cortex, some through the visual. Helen Keller described herself as driven solely by emotions, like a beast, before she had any kind of language: https://www.reddit.com/r/woahdude/comments/1jyo53/hellen_kel...