The AI training and reinforcement learning scoring rewards AI that get more right even when they guess. This is a common strategy in multiple choice tests. In business humans learn to express levels of uncertainty when they give answers. I think the answer might be A but I am not sure. It is maybe 20% ...

Read more

---

Link :
https://www.nextbigfuture.com/2025/09/openai-research-ai-hallucinations-is-strategic-guessing.html