OpenAI has launched GPT-5, a more advanced AI model for ChatGPT that excels in math, coding, writing, and healthcare advice. According to OpenAI, GPT-5’s hallucination rate has decreased to 9.6%, a significant improvement from GPT-4’s 12.9%. The new model boasts 44% fewer responses containing major factual errors, emphasizing its reliability. However, about one in ten responses may still include inaccuracies, raising concerns, particularly in critical fields like healthcare.
While GPT-5 shows improvements, researchers note that hallucinations remain an intrinsic challenge for large language models (LLMs). Evaluations indicate that GPT-5 performs better when it can access the web, significantly reducing hallucinations. Nevertheless, without web access, GPT-5 demonstrated higher error rates during fact-checking tests. Users should verify information through web searches, especially for significant queries. Despite advancements, instances of inaccuracies have already surfaced, highlighting the ongoing challenges in AI accuracy.
To try GPT-5, follow specific steps on OpenAI’s platform.
Source link
