[xuchenglan]

Interesting angle. I’m curious how you’re defining ‘what AI thinks’ in a measurable way. If you’ve explored any concrete evaluation methods, I’d be interested to hear more.