A model can connect three words creatively because the words already identify the concepts to connect. Three images demand an additional step: discovering which concepts matter.
CREATE's October 13 paper uses visual Remote Associates Test problems, with human validation and evaluations of eleven vision-language models. Its findings suggest that associative strength depends on textual cues and does not transfer uniformly to visual puzzles. Read the research paper.
The caption can give away part of the task
Supplying cue concepts changes the problem. It can remove ambiguous visual interpretation and select the intended association before the model begins reasoning.
That makes paired image-only and text-assisted tests valuable. A large gap identifies a boundary between finding the relevant visual concepts and combining concepts already provided.
The small human study and selected puzzle set limit claims about creativity as a whole. The benchmark measures a particular form of association, not artistic worth or general originality.
For design assistants, inspect where the creative leap occurred: in the model, in a captioning pipeline or in the user's wording. A fluent explanation of a connection is less informative when the input has already supplied the hard part of that connection.
References
Questions, news tips or corrections? Contact [email protected].
