Last week, Openai released its new O3 and O4-Mini Reasoning models, performing much better than their O1 and O3-Min predetiles and have new abilities such as adding AI devices to “thinking with image” and “thinking with images” and more complex results.
However, according to Openai’s internal trials, these new O3 and O4-MINI logic models also have much frequent hallucinations compared to the previous AI models, reporting Techcrunch. This is unusual because new models improve underlying AI technology as less hallucinations.
LLMS and Reasoning AIS, a “hallucinations” occurs when the model makes information that seems confident but really no effect. In other words, when you ask the question questions, it can answer with an answer that is false or wrong.
Openai’s in-house benchmark persona-used to measure the factual accuracy of its AI model while talking about people-on the fact that the O3 halts in 33 percent of the reactions while O4-Mini has upgraded 48 percent. Comparatively, the old O1 and O3-Mini models respectively performed 16 percent and 14.8 percent hallucinations.
So far, Openai says they do not know why the new argument model has increased hallucinations. The hallucinations for creative efforts may be cured, but they reduce the reliability of AI assistants such as chatgip when used for tasks where accuracy is paramount. In a statement by Techcrunch, an Openai representative said the company is “working to improve continuously [their models’] Accuracy and reliability. ,
This article originally appeared on our sister publication PC Four Alla and was translated and localized from Swedish.
