ChatGPT can be a little sneaky. Recently asked if it’s ever messed up when helping physicians make clinical decisions, it only came clean with examples of stumbles after being pressed on the matter.
The study's authors communicated with ChatGPT-4 in both English and Japanese, noting that it appeared to struggle more with the complexities of the Japanese language.
In healthcare, this typically manifests as lingering regret after organizational constraints prevent physicians from acting in the best interest of their patients.