This is part-bug, part-feature request. Following on from a response by Zac on Discord explaining how "the thinking boxes are native HTML <details> tags", and that "when their content is focused (via the native page text search) the browser expands them", I would like to request that this be changed.
When searching through assistant responses in long threads, and particularly when using thinking-heavy models, it can be quite difficult to find the information the assistant had previously outputted (being the information that I read and am therefore looking for via command + f). The thinking itself clogs the search interface for me. I understand that the current HTML implementation is "not necessarily intentional / by design, just a consequence of ... not using custom elements for these things", so this would be a great improvement!
