Might be that the training set is too old; “That one with the fat dragon” got me Kung Fu Panda, because Po, while not a dragon, was fat.
Similar prompts (see my other replies to the OP) for older films seem to work, so Honour Among Thieves might be to recent. (Which would highlight one of the main issues with these large language models: they have to be periodically retrained with new data for them to remain accurate and practical.)
OK, last last one; “That one with Bill Murray and the gopher” got me Caddyshack, which I haven’t seen but does seem to be a better match than Groundhog Day, which is what I was going for (trying to see if the AI could figure out I was mixing my rodents; turns out there are too many Bill Murray rodent focused films for that to be a valid test, my bad).
I had to do it in a third browser 'cause the login prompt came back, though, so I won’t be doing any more attempts, between that and the GDPR prompt it’s pretty clear they don’t want people using their site. Shame, as the tool seems pretty good, but what can you do. 🤷♂️
OK, opening it on another browser worked. “why it was so easy to blow up the death star” got me Rogue One. So far 3/3, plus a login prompt which drops the score to 0/4.