It is a limited study, but it does show the gap that a novice may miss. This will be closed as the companies can retrain and isolate domains and get better at picture (figure) descriptions. Probably not a great idea long term since the analysis and implications is where humans shine, but as a first pass to get the boilerplate down, it may do a reasonable first pass.
https://www.theregister.com/2023/06/08/scientists_ai_recognition/