alt Hacker NewsIt probably gives them confidence that they can accurately see a thing even though they don't know what that thing is.
I could also imagine a lot of safety around leaving things outside of the current task alone so you might have to bend over backwards to get new objects worked on.
There is no such thing as "thing" here.
These models are trained such that the given conditions (the visual input and the text prompt) will be continued with a desirable continuation (motor function over time).
The only dimension accuracy can apply to is desirability.