OpenAI’s recently launched GPT-5 is facing significant issues with “hallucinations,” where the AI generates inaccurate information. While OpenAI claims improvements in accuracy compared to earlier models, users and experts indicate that hallucinations remain a problem. A Futurism report highlighted one instance where GPT-5 inaccurately stated Poland’s GDP as over $2 trillion, while the IMF cites it at around $979 billion. Discussions on Reddit confirm these concerns, with users reporting substantial errors despite the model’s high confidence. OpenAI’s blog attributes this issue to current evaluation techniques that reward confident, yet factually incorrect responses. Experts from the Walter Bradley Center critique GPT-5’s reliability, warning against its use in critical contexts. OpenAI admits these challenges, suggesting future training should focus on recognizing uncertainty instead of providing plausible-sounding assertions. Until improvements are made, GPT-5’s tendency to produce confidently erroneous information persists.
Source link
