[js8]

That's been my experience too. I had some nighttime pictures taken from a plane ride and I wanted Claude to identify the area on the map that corresponds to the photograph.

Claude wasn't able to do it. It always very quickly latched onto a wrong hypothesis, which didn't stand up under further scrutiny. It wasn't able to consider multiple different options/hypotheses (as human would) and try to progressively rule them out using more evidence.