tl;dr: Berkeley researchers trained models on a possibly-accidentally-leaked Chinese dataset -- images of US naval ships, suspiciously labeled with bounding boxes around their radar systems.
They're making the point that training models like this can give some intelligence towards another country's ML capabilities (here, China). In my opinion, a cool work that has way more "contribution" than your average AI conference paper.
(And full disclosure: I contributed a bit of feedback on their writing.)
That's really strange and incredibly frustrating - but slightly less so if it's consistent with all of the bars (including their own).
I take issue with their choice of bar ordering - they placed the lowest-performing model directly next to theirs to make the gap as visible as possible, and shoved the second-best model (Grok-1) as far from theirs as possible. Seems intentional to me. The more marketing tricks you pile up in a dataviz, the less trust I place in your product for sure.
They're making the point that training models like this can give some intelligence towards another country's ML capabilities (here, China). In my opinion, a cool work that has way more "contribution" than your average AI conference paper.
(And full disclosure: I contributed a bit of feedback on their writing.)