OpenAI has developed GPT-4, a cutting-edge deep learning model that can process both text and image inputs, and generate text outputs with impressive human-level performance on various professional and academic benchmarks. Unlike its predecessor GPT-3.5, GPT-4 has been trained on an extensive corpus of data to produce coherent and contextually appropriate responses to different inputs, making it more creative, reliable, and capable of handling nuanced instructions. It has undergone rigorous testing on traditional benchmarks for machine learning models, outperforming existing large language models and most state-of-the-art models. ChatGPT and the API currently provide text input capabilities, while the image input feature is being prepared for wider availability through a single partner collaboration. OpenAI has also released OpenAI Evals, a framework for automated evaluation of AI model performance, to encourage everyone to report deficiencies in their models and guide further improvements.
