GPT-4 技术报告
GPT-4 是一个多模态大型语言模型，接受文本和图像输入，生成文本输出。在各个专业和多模态基准测试中，GPT-4 表现出与人类可比的能力和性能，在包括法律、医学、高级数学等广泛领域内的人类考试中都达到了及格分数线以上。GPT-4 在语言理解和推理能力方面相比GPT-3.5有显著改进。
GPT-4 is a large multimodal model (accepting text inputs, issuing text outputs) that, while less capable than humans in many real-world scenarios, exhibits human-level performance on various professional and academic benchmarks, including the simulated bar exam, math olympiad, and medical licensing exam.