OpenAI Unveils GPT-4.5: More Natural, Reliable AI Interaction
OpenAI has rolled out GPT-4.5, the latest iteration of its AI model, to ChatGPT Pro users and developers worldwide. This new version promises a more natural interaction with improved understanding of user intent and enhanced 'EQ'.
GPT-4.5 is now available on web, mobile, and desktop platforms, with a wider rollout planned in the coming weeks. The model has been trained on Microsoft Azure AI supercomputers, resulting in reduced hallucinations and increased reliability across various topics.
In evaluations, GPT-4.5 scored a SimpleQA accuracy of 62.5%, a significant improvement over GPT-4. Its hallucination rate also decreased to 37.1%. Human testers preferred GPT-4.5 in creative intelligence, professional queries, and everyday queries, demonstrating its stronger aesthetic intuition and better understanding of human intent.
OpenAI has focused on scaling up pre-training and post-training, with an emphasis on unsupervised learning, to achieve these advancements.
GPT-4.5, now available to ChatGPT Pro users and developers, offers a more natural and reliable interaction experience. With reduced hallucinations and improved understanding of user intent, it represents a significant step forward in AI development. OpenAI continues to push the boundaries of AI with this research preview.
