GPT-5 Is Making Huge Factual Errors, Users Say
SMRTR summary
A cutting-edge AI chatbot with "PhD-level intelligence" can't seem to get basic facts straight. GPT-5, OpenAI's latest large language model released last month, has been confidently serving up falsehoods that even casual users are spotting.
"The scary part? I only noticed these errors because some answers seemed so off that they made me suspicious," wrote one Reddit user who caught GPT-5 wildly inflating country GDPs. Poland's economy was incorrectly stated as having over two trillion dollars, roughly double its actual value.
In another bizarre test, the AI was asked to label body parts on a possum drawing but instead created an image of cowboys with nonsensical labels, marking a hat as a "shoulder" and a shin as a "fand."
OpenAI acknowledges the problem in a recent blog post, explaining that AI models hallucinate "because they are trained to get things right, even if it means guessing."
Even Sam Altman, OpenAI's CEO, seems to recognize the model's shortcomings as he pivots to hyping the future GPT-6 after the troubled launch.
The takeaway? When AI confidently presents information, a healthy dose of skepticism remains essential.
SMRTR provides this summary for quick context. The original article belongs to Futurism.
Read the original article