Infographic Glitch
As noted in yesterday’s post, the ChatGPT-5 model’s visual generation can drift into bizarre territory. A US map emerged with states labeled “Onegon” and “Gelahbrin,” and a presidential timeline listed only 26 leaders, including fictitious “Willian H. Brusen,” with scrambled dates. Text prompts alone produced accurate lists, showing the flaw lies in text-within-image generation, not data recall. So it is clear this has not improved much in ChatGPT-5, even as Nick Turley says it “should feel smarter starting today.” Read more →